Reflective Teaching of Logo

This article was downloaded by: [University of California, SanFrancisco]On: 16 December 2014, At: 23:00Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number:1072954 Registered office: Mortimer House, 37-41 Mortimer Street,London W1T 3JH, UK

Journal of the LearningSciencesPublication details, including instructionsfor authors and subscription information:http://www.tandfonline.com/loi/hlns20

Reflective Teaching ofLogoRichard Lehrer , Mihwa Lee & Allan JeongPublished online: 17 Nov 2009.

To cite this article: Richard Lehrer , Mihwa Lee & Allan Jeong (1999)Reflective Teaching of Logo, Journal of the Learning Sciences, 8:2, 245-289,DOI: 10.1207/s15327809jls0802_3

To link to this article: http://dx.doi.org/10.1207/s15327809jls0802_3

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy ofall the information (the “Content”) contained in the publicationson our platform. However, Taylor & Francis, our agents, and ourlicensors make no representations or warranties whatsoever as to theaccuracy, completeness, or suitability for any purpose of the Content.Any opinions and views expressed in this publication are the opinionsand views of the authors, and are not the views of or endorsed byTaylor & Francis. The accuracy of the Content should not be reliedupon and should be independently verified with primary sources ofinformation. Taylor and Francis shall not be liable for any losses,actions, claims, proceedings, demands, costs, expenses, damages,

http://www.tandfonline.com/loi/hlns20

http://www.tandfonline.com/action/showCitFormats?doi=10.1207/s15327809jls0802_3

http://dx.doi.org/10.1207/s15327809jls0802_3

and other liabilities whatsoever or howsoever caused arising directlyor indirectly in connection with, in relation to or arising out of the useof the Content.

This article may be used for research, teaching, and private studypurposes. Any substantial or systematic reproduction, redistribution,reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of accessand use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

http://www.tandfonline.com/page/terms-and-conditions

http://www.tandfonline.com/page/terms-and-conditions

THE JOURNAL OF THE LEARNING SCIENCES, 8(2), 245-289 Copyright O 1999, Lawrence Erlbaum Associates, Inc.

Reflective Teaching of Logo

Richard Lehrer, Mihwa Lee, and Allan Jeong Department of Educational Psychology

University of Wisconsin, Madison

A long-term instructional experiment assessed the cognitive consequences of reflective teaching practices, compared to inquiry teaching practices, on the acquisition and transfer of Logo computer programming for 2nd- and 5th-grade students. Inquiry teaching represented previous "best practices" in which teachers elicited predictions, asked leading questions, and assisted students when they encountered programming impasses. Reflective teaching modified inquiry teaching by explicit encouragement of adesign stance where students assumed roles as potential designers of Logo as well as actual roles as the designers of their own programs for peer audiences. Other tools for reflection included writing summaries of their programming experiences, using programming templates, and working with "microworlds" that helped students objectify their experiences with Logo. Multiple levels (syntactic, semantic, schematic, strategic, and beliefs) of Logo knowledge were measured during and after instruction. Between-group differences over repeated measures consistently favored the group participating in reflective instruction. However, rather than simple differences on every measure, the pattern of mean differences over time was most consistent with Mayer's (1985) proposal of achainof cognitiveconseqqences regulating the acquisition of Logo. Moreover, participation in the reflective context facilitated general transfer of specific skills like debugging and summarization for both grades, but no differences in general transfer were observed between instructional conditions for skills not explicitly targeted for instruction. Children participating in the reflective context developed beliefs about programming practices that were tightly coupled with their performances. This tight coupling underscores the reciprocal relation between the social and the individual in the acquisition and transfer of cognitive skill.

"He who offers 'a penny for your thoughts' does not expect any great bargain" (Dewey, 1933, p. 4). From this pejorative preamble, John Dewey went on to ad- vocate reflective thinlung as a paramount educational aim. For Dewey, reflec-

Correspondence and requests forreprints should be sent to Richard Lehrer, Wisconsin Center for Ed- ucation Research, University of Wisconsin, Madison, 1025 West Johnson Street, Madison, WI 53706. E-mail: [email protected]

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

tion had multiple senses, including consideration of assumptions, mental restructuring of events, and "some justification outside of the chain of ideas" (p. 6)-a perspective informed by the philosophical analysis of consciousness artic- ulated by his colleague, George Herbert Mead (1910, 1932). Reflection involved a mental "lifting out" and inspection divorced from the plane of immediate activity; reflection made explicit and symbolic that which previously was enacted tacitly in the situation. Dewey also suggested that reflective habits of mind be cultivated at an early age and that learning environments be designed with this objective in mind. He recognized the importance of communities of practice to the construction of mindfulness, suggesting that although older persons might be oriented to reflect on their experience by professional or other social identities, opportunities for reflection in children would need to be created by explicit ped- agogical intervention.

With the latter admonition in mind, the purposes of this study were threefold. First, we attempted to design an instructional context that would encourage children to engage in reflective thought. An important part of our research was to develop a learning community where reflection would be encouraged, supported, and given fertile grounds for development. Second, we aimed to examine the cognitive consequences of learning in this reflective context, with an eye toward look- ing at the relation between reflection and the development of both simpler and more complex forms of knowledge. Third, we intended to analyze the relation between the social, here represented by a set of communally shared design practices (discussed later), and the individual.

We instructed children in the elementary grades in the Logo programming language for most of a school year. Our rationale for selecting Logo involved several considerations. First, to study reflection in the sense proposed by Dewey, we needed to help children develop knowledge about a complex domain, one where previous research would allow us to effectively characterize student learning. Pre- vious research could also assist the identification of pedagogically fruitful targets for reflection, enabling the instructional design to be well informed. Logo is a much-studied domain where learning clearly encompasses multiple levels and forms of cognitive skill. Mayer (1985, 1992), for example, suggested a hierarchy of Logo programming knowledge: syntactic (e.g., elementary Logo commands), semantic (e.g., what elementary Logo commands do), schematic (e.g., how Logo commands can be orchestrated to accomplish some goal), and strategic (e.g., how program goals can be coordinated by planning heuristics such as program modularization). Other research (Clements & Merriman, 1988; Lehrer & Littlefield, 1993) indicated that these levels of programming skill draw on multiple forms of skill, including working memory resources (e.g., spatial and verbal processing capacities) and representations (e.g., how Logo commands are organized). In sum, there is abundant evidence that learning about Logo involves orchestrating multiple levels and forms of knowledge. Moreover, the research base provides suf-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

REFLECTIVE TEACHING OF LOGO 247

ficient grounds for characterizing this complexity and makes it feasible to track individuals' conceptual change.

A second ground for selection of Logo was that learning to program in this language involves ideas that are readily "taken from the head" and made visible as objects for reflection. The generation of programming code allows children and teachers to visualize a trace of cognitive activity, an important preamble for reflection about that cognitive activity. Moreover, for many forms of programming activity, like debugging, strategies can be decomposed and made visible to children textually or graphically (Carver & Klahr, 1986; Clements, 1990; Lehrer, Guckenberg, & Lee, 1988; Swan, 1989).

A third consideration was that Logo is representative of a design domain (Si- mon, 1969)-a domain in which the nature of the problem is often ill specified and defies comprehensive statement (e.g., "Design a good tutoring program."), and one in which the means to accomplish goals often involve juggling constraints rather than finding optimal algorithms (Lawson, 1990). All of this poses a "paradox and predicament" for learning in that one must do and reflect about what one does simultaneously-a form of "reflection in action" (Schon, 1990, p. 99). Consequently, efforts to promote reflection are especially critical in design domains.

DESIGNING REFLECTIVE TEACHING

Although the theoretical tradition initiated by Dewey (1933) underscores the importance of reflection to education, it is silent about fruitful targets for reflection in specific domains. Consequently, we relied on previous research about programming to determine what might be important to target for reflection. First, studies of expert and novice programmers suggest that experts take a design perspective on programming (they think about the goals of the program and potential relations among chunks of ctde to accomplish their goals), whereas novice programmers tend to see programming as equivalent to writing code (McKeithen, Reitman, Rueter, & Hirtle, 1981; Soloway, Spohrer, & Littman, 1988). Second, and related to the first set of findings, novice programmers often identify programming actions at the level of individual programming statements; they fail to "chunk" program code in relation to the goal it is intended to accomplish (Lehrer, Guckenberg, & Sancilio, 1988). For example, instead of thinking about a program to draw a house as a decomposition of related pieces (e.g., roof and body), novices, especially children, tend to write "spaghetti codew-long strings of statements that eventually accomplish the goal but hide the relations among the statements (e.g., the programming structure is flat rather than hierar- chical). Third, and also related to the first two points, children often treat programs as local constructions without thinking much about their connections to

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

248 LEHRER, LEE, JEONG

other related programs or ideas. This tendency highlights the importance of "bridging instruction" to help children develop relations among programming constructs (Littlefield, Delclos, Bransford, Clayton, & Franks, 1989). Fourth, the very interactivity of programming can lead children to chains of action-reaction, rather than mindful consideration of the consequences of their activity (Cope & Simmons, 1994). These research findings collectively suggest fruitful targets for reflection because more mindful consideration of any or all of them may have a powerful impact on children's learning.

These four research findings guided our development of reflective teaching practices, each of which was intended to promote more mindful consideration of Logo. To this end, we instigated four general classroom practices, each designed to focus on a key target for reflection in the context of programming, as follows: (a) dialogue about design principles and real audiences for program (to promote reflection about Logo programming as design rather than as simply writing code), (b) provision of programming templates (to exemplify and promote reflection about key principles of programming design, like modularity), (c) summarization of the action of programming code (to promote reflection about the action of chunks of code rather than lines of code), and (d) compare-contrast questions (to help children reflect dialogically about relations among different programs or chunks of code, or about relations between programming and everyday experiences). We describe next how we implemented each practice.

Student Designers

Students reflected about design, both at the level of the semantics of individual commands and at the schematic and strategic levels of program design, by engag- ing in teacher-led dialogue about it (see Harel, 1991). To stimulate reflection about the semantics of programming commands, we asked questions like, "If you were making up Logo [if you were one of Logo's inventors], how would you do this? Why? What are some other ways that you might do this?'To illustrate, the instructor might say:

If you were making up Logo and you wanted to tell the turtle to follow its nose so you could get from here to there [instructor points], what word would you use? Why? . . . What should the units be? Why? Why not? Which one should we use?

These questions promoted student dialogue about the assumptions and design principles of the language; students' consensual choices were occasionally incorpo- rated as conventions by modifying Logo to reflect these conventions. For example, second-grade students participating in the study (described later in more detail) decided to measure the angularity of the turtle's turns in units of one complete revolution of their bodies. We subsequently modified Logo to correspond to this change

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


from Logo's convention of turns in degrees. To promote consideration of design strategies at the program level, students' programs were tested periodically by having other students use them. Student observation of users' reactions to their programs was intended to prompt consideration of alternative designs and help students develop a language of design (e.g., "I should have provided some feedback about . . . ."). Developing a language promotes articulation of thinking by helping students understand the origins of their beliefs and how to understand those held by others (Olson & Astington, 1993).

Programming Templates

We provided seven programming templates to students participating in the reflective context. These templates were written in Logo and were displayed on the computer screen whenever the student accessed the template through the keyboard. Each programming template exemplified a major programming construct, such as breaking problems down by writing separate subprograms (i.e., problem decomposition) or using procedures to "package" collections of commands (i.e., program modularity). The rationale for this approach was to objectify (make visible in code) important programming practices and to make it easier for students to compare their practices with those of more skilled programmers. The programming templates were also intended to serve as source analogs for programming solutions. In addition to problem decompositiori and modularization, templates were written for programming conventions about variables, conditionals, list processing, looping, and debugging.

Summarization

To promote reflection about chunks of code rather than individual lines of code, students summarized the actions of their procedures and what they had accomplished each day. Summarization-ski11 training was modeled after that of Palinscar and Brown (1984) and Day (I986), beginning with short passages of text and pro- ceeding to program code. The rationale for this component of reflective teaching was that, by summarizing procedures, children would reflect on the program's purpose (thus bridging the programming actions to the larger context of their use), and by summarizing their accomplishments for the day, children would similarly reflect on the connection between purpose and structure.

Compare-Contrast Questions

We asked students to compare what they were doing to other things that they knew about or asked them to articulate similarities and differences among their procedures. The rationale was to help children develop problem schemata and to bridge

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


from their knowledge of Logo to related contexts. The questions were also designed to help students distinguish necessary commands from superfluous commands when they designed a program. We intended to forestall children's tenden- cies to simply cut and paste a set of commands that accomplished some purpose and package them as a procedure, encouraging instead mindful consideration of opera- tors and a pruning of the problem space.

Objectifying Specific Skills

In addition to the four large-scale components of reflective instruction noted earlier, we also attempted to increase the likelihood of reflection about more specific skills. For example, when teaching children about the syntax of Logo's move and turn commands, we used a program, "Tricky Turtle," that randomly introduced common bugs (like confounding move and turn commands, into student's commands; Lehrer & Littlefield, 1991). Students were asked to predict the commands used by the tricky turtle, and discrepancies between their prediction and the actions taken by the turtle were made visible by drawing different turtle paths-one for their predictions and one for the actual path. Students then described any discrepancies between the two paths with respect to the commands used and their effects. The contrast between the two paths and the two sets of commands served as a concrete object for inspecting and reflecting about the semantics of Logo's turn and move commands. (We further describe these efforts to make specific mental skills more visible in the Method section.)

In summary, we designed a program of reflective instruction by considering previous research about programming, especially research about children learning Logo. Taken as a whole, the research on programming supports the need to promote reflection about programming activity, especially because the very interactivity of computer programming can disguise the virtues of considered action. To counteract this tendency, we designed instruction to foster reflection.

THE TEACHING EXPERIMENT

We decided to test our approach to reflective teaching at two grades, second and fifth, a comparison motivated by theories of development that suggest significant change in children's processing capacity, problem-solving skills, and metacognitive skills during this age span (Case, 1978,1985; Flavell, 1985; Siegler, 1991). Although little consensus exists about the appropriate explanation for transitions in these mental skills, nevertheless there is widespread agreement that these transitions have important implications for instruction. For example, young children often fail to reflect on their thinking spontaneously (Brown, Bransford,

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


Ferrara, & Campione, 1983). It remains questionable, then, whether forms of assis- tance designed to promote reflection will prove profitable for children in the second grade. On the other hand, research in cognitive strategy training suggests that older children especially, and perhaps younger children as well, should be able to profit from appropriately designed instruction to promote monitoring of strategic behavior (Pressley, 1986; Pressley, Goodchild, Fleet, Zajchowski, & Evans, 1989; Symons, Synder, Cariglia-Bull, & Pressley, 1989).

Inquiry Versus Reflective Instruction

To assess the effectiveness of reflective instruction, we considered several potential benchmarks. One possibility was to compare the reflective intervention to a "discovery" condition where children would be left to their own devices to find out how to solve the programming problems posed to them. However, this possibility did not meet the criteria of a useful comparison because children in discovery programming contexts generally accomplish little without teacher mediation (Clements, 1990; Lehrer, Guckenberg, & Sancilio,l988; Littlefield et al., 1988; Fay & Mayer, 1987). Among potential forms of teacher-mediated contexts for learning Logo, previous research suggested the utility of teacher-guided inquiry-asking students leading questions (e.g., "Why do you think that worked this way?')), eliciting predictions (e.g., "What do you think will happen next?')), and as- sisting students to resolve impasses (Lehrer, Guckenberg, & Lee 1988; Lehrer, Guckenberg, & Sancilio,l988; Littlefield et al., 1989). In many ways, inquiry-based instruction helps children become more "mindful" (Salomon & Perkins, 1989) about their computer programming, in part because teacher questions increase opportunities for student reflection. The reflective instruction was designed to improve systematically on the inquiry-based instruction by increasing children's opportunities for reflection in the manner described previously. Conse- quently, we decided to contrast reflective instruction with our previous best efforts rather than a less valid control condition. As we will describe more fully, our comparison of these designs for learning embraced three dimensions: (a) learning Logo, (b) transfer, and (c) the relation between socially constituted practices and individual performances. Collectively, these contrasts afforded a "design experiment" (Brown, 1992).

Learning Logo: A Chain of Cognitive Consequences

The contrast between reflective and inquiry-based instruction suggests a pattern of expected outcomes, rather than simple differences between groups for all types of learning because the acquisition of the four levels of Logo knowledge (syntax, semantics, schematic, strategic design) follows a chain of cognitive consequences

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


(Fay & Mayer, 1988; Mayer, 1992). The cognitive-chain hypothesis suggests that the acquisition of higher levels of programming skill, such as those involved in program design (strategic knowledge), relies on the consolidation of lower levels of knowledge, such as one's knowledge of Logo's commands (syntactical knowledge) and how they work (semantic knowledge). Acquisition of the four levels of programming knowledge can proceed in parallel, but consolidation of these levels proceeds serially, from lowest to highest. This order of consolidation is consistent with several theories of skill acquisition. For example, in Anderson's (1983, 1993) model of the adaptive character of thought, learning how to use Logo commands involves the formation of simp1t:r and fewer production rules (if-then clauses) than learning how to design Logo programs. Similarly, Powers' (1973) control theory of behavior suggests that the information integration of higher order control principles such as those involved in program design (the "principle" level of control) nec- essarily take longer periods of' time than that involved in developing standards to control Logo's syntax.

In this study, the cognitive-chain hypothesis suggests a pattern of outcomes rather than uniform differences between instructional contexts on all measures of learning. First, because the reflective teaching is aimed at thedevelopment of higher levels of knowledge about Logo (semantics, schematic, strategic), no differences between groups are expected for measures of syntax. Second, reflective practices ac- celerate the acquisition of higher levels of knowledge. Consequently, at any particular time of measurement, the prediction is that there will be no difference between groups for consolidated knowledge, but that there will be differences between groups further up thechain (at a higher level) due to the benefits ofreflection. For example, early in the instructional sequence, one might expect differences between instructional contexts only at the level of semantics because consolidation of syntax is not expected to be affected by the difference in instructional contexts. Be- tween-group differences would not be expected higher in the chain (e.g., schematic, strategic) because these presumably rely, in turn, on consolidation of semantics. Later in the instructional sequence, one would expect differences between groups to center at higher levels of knowledge (e.g., schematic, strategic), given consolidation of lower levels. Overall, the predicted pattern is like a sliding scale: The advantages of reflection for learning any particular level of knowledge are gradually washed away by opportunities to learn in the inquiry context, so the differences between groups appear higher and higher on the chain. Perhaps, with enough time, no differences between instructional conditions would be found at any level of Logo knowledge. However, in this study, instructional time was not unlimited, so a time lag is predicted: Semantic, schematic, or strategic knowledge is consolidated first by the reflective group and then by the inquiry group.

Differences between grades are also governed by the chain of consequences. Younger children, due to their relative paucity of cognitive resources (i.e., working memory) and perhaps lower ability to take advantage of reflective opportuni-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


ties, are expected to proceed more slowly along the chain of cognitive consequences. Consequently, early in the instructional sequence, between-group differences might occur for younger children at the semantic level (indicating early consolidation of the syntactical level) and for older children, at the schematic level (indicating faster consolidation of the syntactic and semantic levels).

In accordance with our notions of the importance of reflection in complex domains, assessments were designed to measure multiple levels of Logo knowledge (i.e., syntactic, semantic, schematic, strategic), and these levels of knowledge were sampled twice-after 1 month of instruction and at the conclusion of the instruction. The multiple times of measure allowed us to test the prediction of a time lag in between-group differences among the four levels of knowledge about Logo. As we describe later, we used multiple measures of most forms of Logo knowledge in an effort to measure each level adequately; we had tested the measurement properties of most of the measures previously (Lehrer, Guckenberg, & Lee, 1988; Lehrer & Littlefield, 1993).

Transfer

We compared transfer of learning in the inquiry and reflective contexts after 5 months of the school year. The comparatively long duration of instruction was necessary because we were interested in potential differences between instructional contexts with respect to complex as well as simpler forms of knowledge-and these take long periods of time to acquire. Moreover, it is fruitless to ask questions about transfer if programming skills are not mastered (DeCorte, Verschaffel, & Schrooten, 1992). To this end, we examined between-group differences on measures of general and specific transfer. Traditionally, programming has been viewed as a vehicle for the transfer of general problem-solving skills and even as an aid to creativity (Clements, 1986), but there has been an increasing consensus that specificity of both components and processes is the key to transfer-this teaching experiment provides one more test of this proposition (e.g., see Detterman & Sternberg, 1993; Singley & Anderson, 1989). In this study, we examined the transfer of specific skills like summarization and debugging as well as more general skills, like planning.

Social-Individual Relations

Consistent with Dewey's emphasis on the importance of communities of practice to the growth and development of reflective thought, we attempted to characterize the relation between children's beliefs about programming practices, a measure of their "internalization" of socially constituted practices (e.g., those derived from their talk about design), and children's programming performances. If practices developed in the reflective context are important to the development of higher levels

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


of Logo knowledge, then one should expect strong relations between children's beliefs about these practices and their level of skill.

METHOD

Participants

Twenty-four second-grade (mean age = 7 years, 8 months; 13 girls and 1 1 boys) and 26 fifth-grade(mean age= 10 years, 6 months; 12 girls and 14 boys) children participated. At each grade, children had the same classroom teacher, with theexception of one student in the fifth grade who had adifferent teacher. (Children in intact classes were assigned randomly to the two teaching contexts, as described later.) All children attended an elementary school located in a suburban school district. The occu- pations of their parents included farming, light industry, and the professions.

Half of the second- and 42% of the fifth-grade students had a personal computer at home. All of the children at both grade levels had previous experience with drill-and-practice software and with arcade-like game software (in shopping malls and/or at home). Most also had prior experience with word processing (58% at Grade 2 and 77% at Grade 5). Thus, the students participating in this study all had some previous experience with computers. However, none of the Grade 2 and only three of the Grade 5 students had prior experience (and that was minimal) with any type of computer programming.

Procedure

Prior to instruction, all participants were interviewed about their previous experience with computers, both in the school and at home. Three tasks measuring verbal and spatial working memory were administered. Scores obtained on the three measures of working memory were standardized within each grade; a sum of these three measures was restandardized to form a general working-memory variable. This working-memory standard score was then combined with a reading-comprehension standard score (obtained from the Iowa Test of Basic Skills in the second grade and a locally developed measure in the fifth grade) to form a general composite of working memory and reading comprehension. The construction of this blocking variable (the composite of working memory and prior achievement) drew from previous studies that indicated strong relations among Logo learning, prior achievement in school (roughly measured by reading skill), and individual differences in working memory (Geva & Cohen, 1987; Lehrer & Littlefield, 1993).

Participants were rank ordered within each grade on the general composite score and assigned randomly by alternate ranks to one of two instructional conditions (Maxwell, Delaney, & Dill, 1984). (This procedure results in a type of ran-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


domized block design.) In first condition, children participated in an inquiry-based, teacher-mediated context where the teacher asked questions (e.g., "What do you think will happen if?'), helped students when they encountered programming obstacles, and generally assisted student performance. In the second condition, children participated in the context designed to increase the likelihood of student reflection. In short, the inquiry-based context was founded on previous best practices, and the reflective context included all of the previous best practices as well as new practices designed to provide children with greater opportunities to reflect on their programming activity.

Second-grade students in both instructional conditions worked for 30 min each day, 4 days each week, for 15 weeks and a total of 52 instructional sessions spanning 5 months of the school calendar (Fall 1989-Spring 1990). Fifth-grade students met for 45 to 50 min each day, at first for 3 days each week and then daily for a total of 55 instructional sessions, also spanning 5 months. The order of the meeting times was counterbalanced between instructional conditions at each grade to avoid teacher-practice effects (i.e., the second instructional session might be more practiced by the teacher than the first instructional session, so the order of instruction -inquiry, reflective4hanged every 4 weeks). The first author was the primary instructor for all conditions, but the second author taught the classes whenever the first author could not. Learning was assessed periodically throughout the course of instruction and immediately following the end of instruction. All measures were administered individually by the second and third authors and by a graduate assistant.

Apparatus

Each student worked with his or her own computer at both grades and instructional conditions. In the second grade, each child used LogoWriter 2.0 running on an Ap- ple IIgs with 5 12K RAM and an RGB color monitor. Instruction took place in the school's Apple computer lab. In the fifth grade, children used LogoWriter 1.1 running on an IBM PC Jr. with 256K RAM and a color monitor. Instruction took place in the school's IBM computer lab. LogoWriter 1.1 on the IBM was modified to make it functionally identical to Logowriter 2.0 on the Apple.

Description of Instruction

At the second grade, the programming topics ranged from a simple introduction to graphics commands (commands to move and turn the turtle) to the design and production of a Logo "moviem-an animated display with sound and text that told a story. At the fifth grade, more complex programming constructs like variables were introduced, and fifth-grade students faced more complex challenges of design, like those involved in creating interactive programs (e.g., Madlibs-a familiar chil-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


dren's game in which children generate lists of category words that get inserted into unrelated stories to great comic effect) as well as games or tutorials for other students. At both grades, the programming skills ranged from simple to complex, with increasing orchestration of skills required over the course of instruction. The primary difference between the two instructional contexts was the four practices designed to increase students' opportunities for reflection, described previously: (a) teaching from a design perspective, (b) student summaries of procedures and daily activity, (c) availability of programming templates, and (d) greater emphasis on compare-contrast questions that bridged between different programming episodes or between programming and other activity. In addition to these four practices, we also introduced ways to objectify and make visible particular skills throughout the course of instruction. These included computer programs and role-playing skits (e.g., turtle drivers who directed other children acting as turtles) to help children reflect about the semantics of move and turn commands, the semantics of looping (i.e., REPEAT), the steps involved in debugging, and the operation of variables in programs.

A sample lesson is displayed in the Appendix for teaching about the syntax and semantics of the REPEAT command (a command for iteration). Differences between the inquiry and reflective contexts are indicated by italics, with comments about the functions of different components of the reflective context displayed in brackets. A summary description of a programming template for modularization is also provided in the Appendix; the description is only approximate because the version for the children was interactive and on the computer. Nevertheless, the use of the programming template as an inspectable object that exemplified key ideas about a concept like modularization should be apparent from inspection of this portion of the Appendix. Lehrer, Littlefield, Wottreng, and Youngerman (1993) contains a complete description of all lessons and all software used at both grade levels. Sample programs representative of second- and fifth-grade program designs are displayed in the Appendix; these programs are representative of the practice of modularization but are briefer than many of the programs students designed due to space limitations. As is typical in design settings, children observed other children's solutions to problems of design and occasionally appropriated pieces of them. This practice was regulated by a classroom norm of explaining the working of code, whatever its source.

Overview of Measures

As indicated previously, we aimed to examine student acquisition of four levels of Logo knowledge, transfer of this knowledge, and the relation between socially constituted beliefs and individual performance. Table 1 displays the wide range of constructs and measures employed in this study. Four measures of working memory were administered because previous research suggested that working memory was

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


TABLE 1 Summary of Constructs and Measures for the Teaching Experiment

Theoretical Construct Measure

Cognitive resources

Logo knowledge: Syntactic Logo knowledge: Semantic

Logo knowledge: Schematic

Logo knowledge: Strategic

Transfer: Specific skills

Transfer: General skills

Attitude Beliefs

Mr. Cucumber Spatial span Sentence span Logo working memory Command recall Move and turn mastery Programming constructs mstery Command triads Procedure triads Cued recall of Logo commands Maze navigation Graphics design problems (two measures) Debugging problems Error detection Text summarization Integrating old-new information Planning Pattern induction Attitudes toward Logo Programming preferences Questionnaire about programming practices

an important source ofindividual differences governing what children learned about Logo. As mentioned previously, the first three measures of spatial and verbal working memory were combined with reading achievement to characterize individual differences prior to instruction. A measure of Logo working memory was administered after 1 month of instruction to verify that individual differences in general forms of working memory were also evident in the domain (i.e., Logo). During and immediately after instruction, we administered 10 measures of Logo learning, en- compassing all four levels of Logo knowledge (syntactic, semantic, schematic, and strategic) described by Mayer (1985,1992). Measures of specific and general transfer were administered at the end of instruction. Measures of attitudes toward and beliefs about programming weredeveloped to assess belief-behaviorconnections. All measures are described next, in order of the construct measured.

Cognitive Resources

Four tasks assessed working memory. Previous research indicated that these four tasks collectively measured individual differences in working memory (Lehrer & Littlefield, 1993).

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

Mr. Cucumber. For Mr. Cucumber (Case, 1985), task administration and scoring was standardized by a Hypercard stack implementation on a Macintosh computer. In this measure of visual-spatial working memory, children were presented with a figure of a cucumber with distinct body parts (e.g., arms, legs, eyes, ears, etc.). Some of the body parts were color coded. Children viewed the figure for 5 sec, then the figure was masked. After a 2-sec delay, children were shown another figure identical to the first, except that the color coding was omitted. Children were asked to point to the parts of the figure that had been color coded. Trials were arranged in order of increasing difficulty (e.g., one body part to six body parts color coded), with three trials at each level. This task established a span of spatial working memory in terms of how many distinct parts of the figure the child encoded and recalled. The child's score was a weighted sum of the number of trials correctly recalled at each level, with a total possible score of 63; KR-20 reliability was .74 (Stanley, 1971).

Spatial matrix task. The spatial matrix task (Case, personal communica- tion, September 1988) administration and scoring was standardized by a Hypercard stack implementation on a Macintosh computer. The second measure of visual-spatial working memory consisted of a series of 4 x 4 arrays. Children were shown from one to six matrices, each with only one cell shaded, one at a time for 3 sec each. After a 2-sec delay, the child was shown a blank matrix (i.e., with no cells shaded) and was asked to point to the cells that were shaded on the previous matrices. As with the Mr. Cucumber task, trials were arranged in increasing order of difficulty, with a total of five levels of difficulty and four trials at each level. Testing continued until the child incorrectly identified cells on three of the four trials at a given level. The child's score was a weighted sum of the number of trials correctly recalled at each level, with a total possible score of 60; KR-20 reliability was .74.

Sentence span. This measure of verbal working memory was a modified version of the Daneman and Carpenter (1983) sentence span task (Lehrer, Guckenberg, & Lee, 1988). Children listened to a series of sentences and remembered the last word of each sentence. After the completion of the series of sentences, children recalled each last word in the order in which they heard it. For example, after hearing, "John walked to school. Sara ate her lunch," the child would recall "school" and "lunch." Five trials were provided at each level and the levels increased from series of two to series of six sentences. Scoring was a weighted sum of the number of trials correctly recalled at each level. The highest possible score was 100. Test-retest reliability was .90 (Lehrer, Guckenberg, & Lee, 1988).

Logo working memory. A working-memory measure assessed children's ability to integrate simple Logo graphics commands. Children read aloud a series of simple Logo commands, like FD 5 0 RT 9 0 BK 2 0 , and remembered the last

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

command in each series. After reading a number of such statements (e.g., two for Level 2, three for Level 3, etc.), children chose among four illustrations the one that best represented the turtle's path if the turtle followed the last command of each statement, in the order in which they were read (an initial heading of 0" was assumed). For example, if the student read FD 5 0 FD 5 0 BK 9 0 , RT 9 0 FD 7 5 RT 9 0, and RT 4 5 LT 9 0 FD 5 0 , he or she would choose the drawing the turtle would make if it went BK 9 0 RT 9 0 FD 5 0 . There were four levels (paths in- volving recall of two, three, four, and five commands) with five trials at each level, except for the second level, where there were six. Responses were scored as right or wrong, and testing was discontinued when the child missed three or more trials at a level. A weighted score was computed for each child, consisting of the sum of the products of the number correct at each level and the cardinality of the level (i.e., Level 2 x 4 right + Level 3 x 2 right = 14). The maximum possible score was 72; KR-20 reliability was .93 (Lehrer & Littlefield, 1993).

Syntactic Knowledge of Logo

Command recall. Children recalled as many distinct Logo commands as they could and wrote each one on a separate line on an answer sheet provided for that purpose. Each unique command received 1 point. Command key combinations also received 1 point, such as Ctrl-F, and so on. Mention of actions like making procedures, planning steps, and debugging also each received 1 point. Everything else was assigned 0. Interrater reliability was .98 (Lehrer & Littlefield, 1993). There was no maximum score because scoring depended on student recall of a potentially large set of commands (e.g., over 200 commands and constructs are available in Logowriter).

Semantic Knowledge of Logo

Two measures assessed children's understanding of how Logo commands worked. Both measures were adapted from previous research (Lehrer, Guckenberg, & Lee, 1988).

Move and turn mastery. The move and turn mastery test consisted of a fac- simile of the turtle placed on a felt board representing the computer monitor. Stu- dents were asked to demonstrate the action of Logo move (F, B) or turn (R, L) commands. All administrations were individual. The turtle was placed at one of four initial orientations: 0,90, 135, 180. The commands were: F 3 , B 3 , R 1 0 , L 1 0 , R 4 5 , L 4 5 , R 9 0 , L 9 0 , R 1 8 0 , L 1 8 0 , R 1 3 5 , L 1 3 5 , R 2 7 0 , and L 2 7 0 , for a total of 56 items. The order of administration was random with the exception that the first two items were at the 0 heading, F 3 B 3 , to famil- iarize children with the task.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


The task was scored as follows: First, each of the 56 items was scored as right or wrong (maximum possible score = 56). If wrong, the type of error was recorded. The error categories were: (a) undifferentiated move and turn (i.e., a move like F 3 was treated as a command to move and turn simultaneously), (b) confusion of move and turn (either a move command was treated as a turn, or a turn as a move), (c) a right-left confusion (i.e., an R 9 0 was treated as an L 9 O), (d) an argument confusion (i.e., an R 9 0 was treated as an R 4 5), and (d) simultaneous right-left and argument confusions (i.e., an R 9 0 was treated like an L 45). Percentage of agreement between two independent raters for the scoring of children's errors was 93. Disagreements were resolved by consensus. The maximum score for number right on this task was 56.

Programming constructs. The measure of Logo programming constructs consisted of items measuring student understanding of (a) flow of control (12), (b) variables (7), and (c) list processing (5). Items were taken from a longer measure developed previously (Lehrer, Guckenberg, & Lee, 1988). Some items asked students to predict the outcome of a string of code, like Re- peat 4 [FD 1 0 RT 9 0 BK 101. Other items asked students to trace the actions of programs. For example, given a Logo procedure, students used a felt turtle and board to show

To Hexagon MAKE "Side 40 IF :Side > 40 [PR [BIG HEXAGON]] REPEAT 6 [ FD 30 RT 601 END,

what would happen and explained why. (In this instance, the variable does not serve any useful function, so that we could assess literal comprehension of code.) Fifth graders responded to all items (maximum possible score = 24), but second graders responded only to seven of the flow of control items (maximum possible score = 7). Each item was scored as right or wrong and a total score was used as a measure of knowledge of semantics; KR-20 was .94 for the second grade and .96 for the fifth grade.

Schematic Knowledge of Logo

Schematic knowledge refers to the construction of relations among Logo commands or other programming constructs. Three measures of knowledge organization measured programming schemata.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


Command triads. The command triad task consisted of the presentation of three Logo commands or ideas. Students selected whlch two were most alike and provided their rationale. There were 14 items. For example, given three constructs-(a) REPEAT, (b) R 90, and (c) R 90 F 5 R 90 F 5 R 90 F 5-the student might indicate that the first and third elements were most alike because "you can REPEAT the R 9 0 F 5 pattern three times." Each triad was designed to have different surface and "deep structure" explanations. For example, one might group the first two elements together with the explanation that both "begin with R." Each triad was scored on a 0 to 3 point scale, with 0 indicating a surface-level response such as "they both have letters," 1 for associations that could reflect a deeper structure understanding but without an accompanying rationale, 2 for a deeper level association with a partially adequate explanation, and 3 for a deeper level association with an adequate justification. Percentage of agreement between two independent raters was 96. Disagreements were resolved by consensus. The maximum possible score was 42.

Procedure triads. The procedure triad task consisted of the presentation of three Logo procedures or sets of Logo procedures. Students selected which two were most alike and explained why. Second-graders responded to four items (maximum possible score = 8) and fifth-graders to eight items (maximum possible score = 16). For example, in the following procedure definitions-(a) TO SHAPE, F 6 R 90, F 6 R 9 0 , F 6 R 9 0 , END;(~)ToSHAPE, REPEAT 5 [ F 5 R 7 2 1 , END ; and (c) TO BOX, REPEAT 4 [ F 6 R 9 0 I , EN&-the first and second procedures are most alike on the surface because they share the same title. However, at a deeper level of analysis, the second and third procedures are most alike because their action results in a closed path (e.g., a polygon) or because they both use the REPEAT syntax. Each triad was scored on a 0 to 2 point scale, with 0 indicating a surface level response, 1 a deeper level association with a partially adequate explanation, and 2 a deep level association with an adequate justification. Percentage of agreement between two independent raters was 90. Disagreements were resolved by consensus.

Cued recall of Logo. Students were presented with 24 Logo commands such as FD, BK, RT, LT, REPEAT, SETC , IF, and WHEN. Commands were chosen to sample those taught to students, including commands to move and turn the turtle and other graphics commands, list processing commands, variable assignment commands, and commands to control the flow of action in a program. In the second grade, students read each command aloud and then made a list of commands on paper so that commands with "something in common" were most proximate in the list. Then, this process was repeated three more times with different commands serving as cues for recall (the three cues were FD, [ I , and MAKE). The same procedure was followed in the fifth grade, except that the admin-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


BK RT LT

PU

PD

ST

SETC FILL

SHADE LABEL

PRINT

[ I SE FIRST

END

~ r n REPEAT

FIGURE 1 Ordered tree represen- tation of the cued recall of a skilled Logo programmer.

istration was to the entire group (see Naveh-Benjamin, McKeachie, Lin, &Tucker, 1986). Students' responses to this task were then fitted with an ordered tree algorithm (Reitman & Rueter, 1980) as were the responses of an experienced Logo programmer. The ordered tree solution for the experienced Logo programmer (a colleague not involved in this study) is displayed in Figure 1. The double-headed arrows suggest that the chunk is unordered, but unidirectional arrows indicate an ordering within the chunk. We then scored the number of chunks shared by the ordered tree representations of each student with those of the more experienced Logo programmer, ignoring issues of directionality (maximum possible score = 13). Par- tial credit (half a point) was assigned if the student chunk was a subset of the expert's. Percentage of agreement between two independent raters was 84, with disagreements resolved by consensus. Disagreements occurred most often when student chunks shared partial but not complete membership with expert chunks. The maximum score possible was 13.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


Strategic Knowledge of Logo

Strategic knowledge of Logo was defined operationally as knowledge in use+hil- dren's ability to design and implement programming solutions. Three measures of problem solving and one measure of debugging assessed children's use of Logo knowledge.

Mazes. Children saw a computer display of a maze with three landmarks. Students were asked to write a procedure so that the turtle would "visit" each land- mark in a specified order before exiting the maze. Protocols were scored for the presence of procedures (1 pt.), use of modularization (1 pt.), meaningful procedure titles (1 pt.), overall synchronizat~on of the procedures (i.e., their meshing, 2 pts.), and the degree of match between the output of the procedure written by the student and the original graphic displayed. The maximum score was 7. Percentage of exact agreement between two independent raters for each of these five criteria was 95. Disagreements were resolved by consensus. Pearson correlation between raters was .97.

Graphics Design I. Children saw a graphic design presented on 8.5 x 1 1 in. paper. The designs were "boxes," represented by two adjacent squares for the second grade, and "eyeglasses," a graphic representing eyeglasses with squares and line segments. Students designed a program to reproduce the graphic. Protocols were scored as described earlier for the maze, with a maximum score of 7.

Graphics Design 2. Children saw the graphic design displayed in Figure 2, a figure we called the "arch." In the second grade, students wrote procedures to reproduce the graphic and to label it with the words, "This is Greek." In the fifth grade, students also attempted to reproduce the graphic, but we also required that their program ask a user for a color name and colorize a portion of the figure, de- pending on the user's response. Scoring was designed to tap students' use of (a) modularization (0-7 point scale), (b) meaningful titles (1 point), (c) synchroniza- tion among procedures (0-2 point scale), (d) degree of correspondence between the program's output and the illustration provided (0-2 points), (e) proper use of conditionals (1 point, Grade 5 only), and (f) use of variables to get user input (1 point, Grade 5 only). The maximum possible score was 14 for fifth graders and 12 for second graders. Pearson correlation between raters was .96.

Debugging. We developed a computerized measure of debugging skill in Logo. Students were presented eight items in Grade 5 and the first five of these in Grade 2. Each item was presented as follows: (a) First, the student saw an illustra-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


This is Greek

9

FIGURE 2 Intended graphical output for the third-measure program design and problem solving.

tion of the actual action of the procedure and the result intended by the programmer, and then (b) the student revised the Logo program, presented on the computer, to bring about the intended result. For example, Figure 3 displays the actual and intended actions of a program. Each item was scored on a scale ranging from 0 to 3 for easier items (single bugs and procedures) and 0 to 5 for more difficult items (multiple bugs and procedures). The maximum possible score was 32 for Grade 5 and 19 for Grade 2. Percentage of exact agreement between two independent raters was 96. Disagreements were resolved by consensus.

Transfer

Five tasks assessed the generalizability of what students learned. The transfer tasks were designed to assess transfer of skills taught explicitly, like finding errors, and more general skills that one might learn through programming, like planning.

Error detection. In the fifth grade, we used seven items from an error-detection task developed previously by Carver and Klahr (1987; Klahr & Carver, 1988). Each item depicted an expected outcome and the one obtained by following a set of written instructions. The child's task was to identify and repair the error in the instructions. For each item, we assigned 3 points for correct identification and repair of the error, 2 points for identification and partial repair, and 1 point for identification of the portion of the directions requiring change. One point was deducted for extraneous changes to the directions. The maximum possible score was 21. Percentage of exact agreement between two raters was 96.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

REFL.ECTIVE TEACHING OF LOGO 265

In the second grade, we used an error-detection task developed by Lehrer and Littlefield (1993). This error-detection task also depicts an expected outcome and the one obtained by following a set of written instructions. Each of six items was scored according to the guidelines presented earlier. The maximum possible score was 18. Percentage of exact agreement between two raters was 95.

Summarization. Children read six brief passages and then summarized each. Zero points were assigned to the mere repetition of story detail other than a topical sentence, 1 point was assigned to a response that indicated the topic of discussion but not its gist, 2 points to a response that captured part of the gist of the story, 3 points to a response that expressed the main idea clearly, and 4 points to a response that expressed the main idea and had a supporting statement, or that seemed to express a rather abstract main idea that was consistent with the meaning of the passage. The maximum possible score was 24. Pearson correlation between the scores of two independent raters was .94.

Integration of old and new information. We used a task originally designed by Flavell, Green, and Flavell (1985) to measure children's ability to inte-

I don't know what to do with 90 in RECTANGLE.

FIGURE 3 Sample item from the measure of debugging skill. (Top view is actual output, bottom view is intended output.)

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

grate old and new information. This is a general skill related to the high incidence of interactive goals and constraints in programming. That is, modification of programs nearly always involves the need for updating and even reinterpreting old program code in light of new program code. The task consists of 1 1 maps with accompanying two-part directions that are read to the child. The goal for the child was to determine if the directions were sufficient to locate one unambiguously with respect to landmarks on the map. In the first six maps, any ambiguities about location in the first part of the directions were resolved by the second part of the directions. In the remaining five maps, ambiguities are not resolved. The test was to see whether the child recognized the difference between these two conditions. Only in the second case was it necessary to monitor and to integrate old and new states of knowledge. Scoring was identical to that employed in several other studies and resulted in one score for the unambiguous situations (maximum = 12) and another for the ambiguous situations (maximum = 9; for scoring details, see Lehrer, Guckenberg, & Lee, 1988). Percentage of exact agreement between two scorers who did not know participants' identity was 99. This task was administered to second-grade students only.

Planning. A computer program presented students with a number of house- hold chores that needed to be accomplished (e.g., wash the dishes, feed the dog, vacuum). A number of constraints on the order in which the tasks could be accomplished were also presented (e.g., you must vacuum before you dust). Students were asked to indicate the order in which they would accomplish the tasks. If a constraint was violated, the computer informed the student and suggested trying the last step (the one that violated the constraint) again. Two trials were administered at each grade level. Scoring for this task consisted of the efficiency of the solution suggested by the student (measured as the length of the solution path, with shorter paths considered more efficient), and the number of constraints violated summed over the two trials. All measures were collected by the computer as the student responded (for details about scoring and administration, see Lehrer, Guckenberg, & Lee, 1988).

Pattern induction. The pattern induction task had two components: number series and spatial-pattern completion. For the number series, students were presented with a sequence, such as 2, 4, 6, 8; and then asked to select among three choices the number that best completed the pattern. There were 10 items for second graders and 20 for fifth graders. The spatial pattern completion consisted of a sequence of illustrations. Among three or four choices, students chose the pattern that best completed the illustrated sequence. There were three items of this type. For both the numeric and spatial tasks, each item was scored as wrong or right and then summed to yield a total score. The maximum possible score was 13 in the second grade and 23 in the fifth grade.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


Attitudes and Beliefs

Attitude. The measure of attitude consisted of 16 items based on a previous measure developed by Lehrer, Guckenberg, and Lee (1988). Items were statements such as "I like to spend time working with Logo" and "I like Logo more than most subjects." Students indicated their agreement with each statement based on a 5-point scale ranging from 1 (strongly disagree) to 5 (strongly agree). Negatively worded items, such as "Learning Logo is a waste of time," were reflected. Items were summed to a total attitude score. The maximum possible score, indicating perfect agreement with all 16 statements, was 80. Coefficient a was .92 (Stanley, 1971).

Programming preferences. Two items each presented three different Logo procedures for solving a problem (drawing a square and navigating through a maze). The modularity of the procedures varied; some used subprocedures, whereas others did not. Students were asked to choose the procedure that most closely matched the way they would solve the problem. Endorsement of the most modular procedure was scored as 1 and all others as 0. The potential range of scores was 0 to 2.

Beliefs about programming practices. A questionnaire presented statements of programming activities, such as "I usually write a lot of procedures when the problem is big." Students indicated the extent of their agreement with each statement based on a 5-point scale ranging from 1 (describes verypoorly) to 5 (describes very well). Following preliminary analysis, we formed composites of items as indicators of (a) debugging (i.e., "I find that writing a lot of procedures helps me find mistakes," three items in all). (b) program design (i.e., "My programs usually make decisions," four items in all), and (c) use of problem-solving heuristics ("I break problems down a lot," four items in all). The questionnaire was administered only in Grade 5 due to its reading level. The maximum possible score, indicating perfect agreement with all 11 items, was 55.

RESULTS

The primary analytic technique used was analysis of covariance (ANCOVA), with each student's standardized score on the blocking variable (the composite of working memory and reading ability) serving as the covariate. Instructional condition and grade were between-subjects factors. Separate analyses were conducted within grade whenever incommensurate measurement made between-grade comparisons fruitless (e.g., when second-grade versions of measures had fewer or different items). Four sets of results are presented comparing student learning between in-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


TABLE 2 Preinstructional Means and Standard Deviations for Working Memory Measures

by Grade and instructional Context

Grade 2 Grade 5

Instructional Context Instructional Context

Inquiry Reflective inquiry Reflective -

M SD M SD M SD M SD

Sentence span 18.6 7.7 21.4 8.0 32.0 11.5 33.1 18.0 Mr. Cucumber 24.4 9.4 23.3 8.3 42 1 13.9 41.7 13.3 Spatial span 21.6 0.4 18.9 9.6 32.9 14.6 34.4 12.0

structional contexts and grade. First, we examine students' acquisition of levels of Logo knowledge after 1 month of instruction. Second, we present students' acquisition of Logo knowledge (syntactic, semantic, schematic, strategic) at the conclusion of instruction. Transfer of Logo is then presented, followed by examination of relations between socially constituted beliefs and individual performance.

Table 2 displays the preinstructional means and standard deviations on measures of working memory for each instructional condition by grade. Multivariate analysis of variance (MANOVA) confirmed the expected difference between grades, indicating greater cognitive resources for older children, F (3,46) = 13.22, p < .Ol.

Short-Term Acquisition of Logo

Table 3 displays means and standard deviations by instructional context and grade after 1 month of instruction for measures of (a) Logo working memory, (b) recall of Logo commands (syntactic level of Logo knowledge), (c) mastery of Logo turn and move commands (semantic level of Logo knowledge), and (d) each of two measures of strategic knowledge: solving a graphics design problem (boxes in Grade 2, eyeglasses in Grade 5) and solving the maze-navigation problem.

ANCOVA applied to the measure of Logo working memory indicated the expected grade-related differences, F(l,45) = 26.94, MSE = 222.75, p < .Ol , and the blocking variable (composed of working memory and achievement scores) accounted, as expected, for a significant portion of the within-group regression, F(1, 45) = 19.01, MSE = 222.75, p < .01. There were no differences due to instructional conditions at either grade. Hence, age-related differences in general forms of working memory also were evident within the domain of Logo. For the measure of command recall (syntactical level of knowledge), ANCOVA indicated no differences between instructional conditions, as predicted, but the difference between grades was significant; F(1,45) = 20.64, MSE = 18.03, p < .01.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


TABLE 3 Means and Standard Deviations for Working Memory and Levels of Logo Knowledge

After 1 Month of Instruction

Grade 2

Instructional Context

Inquiry Rej7ective -- M SD M SD

Logo working memory 12.5 7.8 11.4 5.0 Syntactic command recall 8.5 5.4 8.5 4.1 Semantic turn-move mastery 23.1 7.5 33.5 10.9 Strategic design problem 2.5 1.7 3.6 1.5 Strategic maze problem 3.2 1.9 3.8 1.9

Grade 5

InstructionaI Context

Inquiry Reflective

M SD M SD

33.6 23.7 34.2 21.1 13.8 4.3 14.2 3.6 37.4 13.7 50.1 4.6 2.2 1.5 5.4 2.3 3.9 2.1 6.8 2.7

ANCOVA applied to the measure of the mastery of move and turn commands (semantic level of knowledge) indicated significant differences between instructional conditions, F(l,45) = 18.84, MSE = 82.72, p < .Ol, but not between grades. We formed a composite of the two measures of problem solving (strategic level) at each grade. ANCOVA of this composite measure suggested an interaction between grade and instructional condition, F(1, 45) = 5.24, MSE = 11.40, p < .05. Separate analyses within each grade indicated no reliable difference between instructional conditions at Grade 2, F(1,21) = 1.77, MSE = 8.90, p >. 19, but a reliable difference between instructional conditions at grade 5, F(1, 23) = 17.48, MSE = 13.74, p c .O 1. The difference between grades on the composite measure of problem-solving performance was significant, t(48) = 2.24, p < .Ol.

In summary, after 1 month of instruction, we found no differences between reflective and inquiry instruction for the acquisition of Logo's syntax: Students participating in either form of instruction learned about Logo's syntax, although fifth-grade students learned more Logo commands on average than did second-grade students. However, students in the reflective context at both grade levels learned more about the semantics of these commands, but fifth-grade students were on average no better than second-grade students. Students in the reflective context also demonstrated superior problem-solving skills (strategic knowledge) at Grade 5 but not at Grade 2, and Grade 5 children were better at solving problems than Grade 2 children. Differences in working-memory resources for Logo were also evident between grades (the measure of Logo working memory). The results show a form of discriminant validity. First, as expected, the reflective intervention made no difference for children's recall of Logo commands-a measure of syntactic knowledge. Second, consistent with the chain-of-consequences hypothesis, the highest level of between-group differences was centered at the semantic level in the second grade and the strategic level in the fifth grade.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


TABLE 4 Means and Standard Deviations of Measures of Long-Term Acquisition of

Levels of Logo Knowledge

Grade 2 Grade 5


Inquiry Rejective Inquiry Rejective --

M SD M SD M SD M SD -

Command recalld L P

Turn-move masteryh Construct ma.~tery~.~ Command triadd

L P

Procedure triadd Cued recalld Graphic design-2e Debugging"

Note. L = late in instructional sequence; P = postinstructional. =Syntactic knowledge. bSemantic knowledge. CMaximum score of 7 in Grade 2 and 24 in Grade 5.

dSchematic knowledge. CStrategic knowledge.

Long-Term Acquisition of Logo

Table 4 displays means and standard deviations for the acquisition of four levels of Logo knowledge: syntax (command recall), semantics (turn-move mastery, Logo programming constructs mastery), schematic (command triads, procedure triads, and cued recall), and strategic (problem solving, debugging). Some of the measures were administered twice: once during the last phases of instruction and again immediately after instruction was over. The successive administrations were designed to provide a window to change over time for selected measures.

Syntactic knowledge. ANCOVA applied to a composite variable of the simple recall of Logo commands across time indicated no reliable differences between instructional methods, F(1,45) < 1, but reliable differences between grades: F(1,45) = 5.46, MSE = 141.13, p < .05. The blocking variable accounted for a significant portion of the within-groups regression, F(l.45) = 6.50, MSE = 141.13, p < .05.

Semantic knowledge. ANCOVA of the mastery of move and turn commands, a measure of semantics, indicated no significant differences between in-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


quiry and reflective forms of instruction at either grade level: F (1,21) = 1.27, MSE = 84.74, p > .27, at Grade 2 and F(1,23) < 1, at Grade 5. The blocking variable at both grades accounted for a significant percentage of the within-group regression for this measure: F(1,21) = 8.02, MSE = 84.74, p < .O1 at Grade 2 and F(l, 23) = 13.19, MSE = 28.63, p c .O1 at Grade 5 . Separate analyses were conducted at each grade level on the second measure of semantics, Logo programming constructs because the fifth-grade version of the measure included items not administered to the second graders. ANCOVA of the measure of programming constructs indicated no reliable difference between instructional conditions: F(1,21) = 1.88, MSE = 1 . 6 6 , ~ > .18, for Grade 2 and F(l,23) = 3.33, MSE= 8.82, p > .08, for Grade 5. The effect of blocking was not significant at either grade level for this measure.

Schematic knowledge. Students' schematic knowledge was represented by a composite standard score consisting of the sum of standard scores within each grade for the postinstructional administration of command triads, procedure triads, and the cued recall of Logo commands (fitted by the ordered tree algorithm). ANCOVA suggested reliable differences between instructional conditions at the second grade, F(1,21) = 10.68, MSE = 5 . 3 3 , ~ < .01, but not at the fifth grade, F(l, 23) c 1. The blocking variable did not account for a significant portion of the within-groups regression at the second grade, but it did at the fifth grade, F(1,23) = 5.79, MSE = 7.58, p < .01.

Strategic knowledge. Strategic knowledge was assessed by designing a program to reproduce a graphic (the arch) and by debugging. Different forms of these measures were administered to each grade, so the analyses are presented within grade. ANCOVA of the measure of program design suggests reliable differences between instructional conditions at each grade level: F(1,21) = 13.99, MSE = 8.85,p< .01, at Grade 2 and F(1,23)= 11.64, MSE = 1 7 . 2 4 , ~ ~ .01, at Grade 5. The bloclung variable was unrelated to performance on this measure. ANCOVA of the measure of debugging also indicated reliable differences between instructional conditions at each grade level: F( l,21) = 18.68, MSE = 3.7 1, p c .Ol, at Grade 2 and F(l.23) = 23.87, MSE = 35.91, p < .O1, at Grade 5. The blocking variable also accounted for a significant portion of the within-group regression at each grade level: F(l,21)=5.18,MSE=3.71,p<.04,atGrade2andF(l,23)=6.08,MSE=35.91,p < .04, at Grade 5.

In summary, after a prolonged period of teaching and learning, there was no difference in student learning between instructional conditions with respect to either Logo's syntax or its semantics, although fifth graders learned more on average about both forms of knowledge than did the second graders. Students in the second grade participating in the reflectwe context developed more schematic knowledge than their counterparts participating in the inquiry context. No such advantage was found for fifth-grade students. Students participating in the reflective context at

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

TABLE 5 Means and Standard Deviations for Measures of Transfer, Attitude, and Beliefs

Grade 2 Grade 5


Inquiry Reflective Inquiry ReJective -- M SD M SD M SD M SD

- - - - -

Error detection 4.5 3.9 11.1 4.5 9.6 7.1 139 3.1 Summarization 6.5 3.1 11.2 2.7 10.4 4.5 13 9 3.5 Integration Old-New 3.6 2.8 6.2 2.5 - - - - -

Planning Violations 4.1 2.4 3.0 1.4 2.6 2.5 1 9 1.6 Path length 21.3 1.5 21.6 1.7 19.9 2.8 20 5 3.2

Patterns Numeric 5.7 2.5 6.8 2.3 12.4 2.9 12.8 3.6 Spatial 1.1 0.8 1.4 0.8 1.6 0.8 1.9 1.0

Attitude 65.3 8.4 59.8 12.9 58.8 15.9 63.3 13.9 Beliefs

Debugging - - - - 10.4 2.3 12 4 2.6 Problem heuristics - - - - 12.8 2.4 16 8 2.7 Program design - - - - 11.4 2.9 148 2.2 Program preference 0.8 0.7 1.7 0.7 1.2 0.8 1 7 0.5

both grade levels were better able to put their knowledge to use, as indicated by their performance on the measures of program design and debugging. The pattern of results conformed to the cognitive-chain hypothesis-by the end of instruction, lower levels of knowledge were consolidated in both instructional contexts. At the fifth grade, the benefits of reflection were evident only at the top of the chain (strategic knowledge), whereas for younger students, the benefits of reflection started at the schematic level at this second time of assessment.

Transfer

Table 5 displays student performance on measures of transfer by grade and instructional condition. The pattern of results for transfer mimics that of other studies of Logo; transferable skills were those explicitly addressed during the course of instruction. ANCOVA of the measure of error detection in text directions indicates reliable differences between instructional conditions at each grade: F(1, 21) = 14.46,MSE=17.24,p<.O1,atGrade2,andF(1,23)=4.84,MSE=21.81,p<.05,at Grade 5. The correlation between student performance on the debugging in Logo and the error-detection task was substantial at each grade: 422) = .77, p < .05, and

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


r(24) = .60, p < .05, for Grades 2 and 5, respectively. ANCOVA of the transfer measure of summarization also suggests reliable differences between instructional conditions at each grade: F(1,21) = 25.22, MSE = 4.88, p < .01, for Grade 2 and F(1, 23) = 4.64, MSE = 17 .17 ,~ < .05, for Grade 5. Similarly, ANCOVA of the measure of integration between new and old information (given at Grade 2 only) indicates a reliable difference in favor of the reflective condition at the second grade: F(1, 21) = 5.72, MSE = 7.16, p < .03. The composite blocking variable did not account for a significant portion of the within-groups regression for this measure (F < 1). As expected, there were no differences between instructional contexts at either grade level for the more general transfer measures of planning ( F < 1) or inducing numeric and spatial patterns (F < 1). In summary, the pattern of results for the measures of transfer conformed to a now familiar form: When general skills are taught and assessed specifically, transfer is evident. Otherwise, transfer is not evident. The sole exception was transfer of slull in determining the relation between new and old information in the second grade. This finding replicates that of other studies that contrast children participating in well-crafted instruction with Logo to those in other instructional contexts (Lehrer, Guckenberg, & Lee, 1988).

Attitude and Beliefs

Table 5 also displays means for postinstructional measures of attitude toward and beliefs about programming (i.e., preference for modular procedures, and the questionnaire about programming practices).

Attitude. No differences between grades or instructional conditions ( F < 1) were detected for student attitude toward their programming experiences, with the mean judgment centered around the agree (fourth) point of the scale. This finding suggests that students in both conditions and grades were interested in learning Logo.

Beliefs. The measure of programming preference suggests a higher endorsement of modular programming practices (using procedures and subprocedures) in the reflective group at each grade level, although the difference was reliable only at the second grade F(l,21) = 11.08, MSE = 0 . 3 4 , ~ c .Ol.

Analysis of students' beliefs about other programming practices (the questionnaire) was confined to the fifth grade. Students in the reflective condition were more likely to report agreement with statements reflecting ideal practices about debugging, problem-solving heuristics, and program design; for example, "I find that writing a lot of procedures helps me find mistakes" (debugging, t(24) = 2.08, p < .05), "I break problems down a lot" (problem-solving heuristics, t(24) = 4.00, p < .01), and "My programs usually make decisions" (program design, t(24) = 3.23, p < .01). (The tests are of the composite scores for each scale.)

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

TABLE 6 Correlations Among Beliefs and Levels of Logo Knowledge in Grade 5

Variabfe 2 3 4

1 . Heuristics beliefs .68 .65 .48 2. Program design beliefs - .45 .62 3. Program debugging beliefs - .34 4. Cued recall (schematic) - 5. Graphics design-2 (strategic) 6. Debugging (strategic) 7. Turn-move mastery (semantic) 8. Command recall (syntactic)

Note. r > . 3 2 , p < . 0 5 ; r > . 4 2 , p c : . 0 1 .

TABLE 7 Principal Components Analysis of Correlation Matrix of

Beliefs and Knowledge

Variable

Component

1 2

Problem solving beliefs Program design beliefs Program debugging beliefs Cued recall Graphics design-2 Debugging Turn-move mastery Command recall

Correlations among these three belief subscales (about practices related to debugging, heuristics, and program design) and measures of Logo knowledge taken at the conclusion of instruction are displayed in Table 6. The measures of Logo knowledge sample both lower level knowledge, as indicated by command recall (knowledge of syntax), turn-move mastery (knowledge of semantics) and higher level schematic and strategic knowledge, indicated by measures of cued recall (schematic knowledge), designing the arch (strategic knowledge), and debugging (strategic knowledge).

A principal components analysis of the correlation matrix is presented in Table 7. A two-component solution, with varimax rotation, accounted for 68% of the variance. Variable-component correlations of less than .5 are omitted. Inspection of the variable-component correlations indicate clearly that beliefs about programming practices were associated with schematic and strategic knowledge but not

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


with knowledge of syntax and semantics. That is, individual differences in beliefs about the nature of programming corresponded well with how knowledge about Logo was organized and put to use. These individual differences had no relation to what children learned about the syntax or semantics of Logo.

In summary, although students in both grades and instructional contexts were positively disposed toward their learning experiences, children participating in the reflective context developed different beliefs about programming practices than did their counterparts in the inquiry context. At the second grade, children in the reflective group preferred modular programming styles, whereas fifth graders endorsed the utility of programming practices designed to increase program modularization and interactivity.

DISCUSSION

Children in the second and fifth grades worked on avariety of programming and design problems for several months in either an inquiry or reflective instructional context. The inquiry context represented former, research-tested best practices to help children learn Logo. The reflective context was designed to improve on these practices by providing increased opportunities for student reflection. General instructional methods like the adoption of a student-designer approach and the use of programming templates were intended to promote the growth of reflection across a wide range of computer programming activities. Other instructional methods made specific elements of programming more visible as objects for reflection; for example, children role played the turtle to learn move and turn commands and used a microworld to extend the semantics of the REPEAT command. Consequently, our goal was not to instruct individuals about ways to think about their thinking, but rather to design a learning environment that would help make thinking more visible and communal. In each instructional context and grade, we assessdd children's learning of (a) multiple forms of Logo knowledge (syntactic, semantic, schematic, and strategic), (b) transferable components of this learning, and (c) children's beliefs about and attitudes toward their learning experiences. At each grade level, we found consistent and persistent differences between the inquiry and reflective instructional contexts with respect to (a) what children learned about Logo, (b) transferable aspects of this learning, and (c) children's beliefs about their learning experiences. However, rather than simple mean differences between instructional contexts, the results on most measures conformed to a pattern suggested by a chain of cognitive consequences.

A Chain of Cognitive Consequences

The pattern of mean differences between instructional contexts among the measures of the four levels of Logo knowledge supports a chain of cognitive conse-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


quences, in which learning of lower levels of programming knowledge supports the acquisition of higher levels of competence (Fay & Mayer, 1988; Linn, 1985). In this study, lower levels of knowledge were embedded within higher levels of knowledge, as suggested by theories of cognitive apprenticeship that caution against learning skills in isolation (Collins, Brown, & Newman, 1989), so the claim is not that one level must be mastered before learning anything about the next. In- stead, the claim of the theory as applied to this study is that lower levels of learning are consolidated before higher levels are consolidated.

Turning now to the pattern evident in the data at Grade 2, after 1 month of instruction in the second grade, no difference between instructional conditions was evident at the lowest level of knowledge (syntax), suggesting that this skill was consolidated at the time of measurement. However, students in the reflective context knew more about the semantics of Logo, so the prediction of the interaction between reflective practices and the chain of cognitive consequences was confirmed. The absence of between-group differences for designing programs was interpreted best as evidence of the difficulties students in both contexts experienced in orchestrating Logo commands to solve programming problems, rather than as evidence of consolidation or mastery after only 1 month of instruction. Assuming that Logo's syntax had been (relatively) consolidated after 1 month of instruction, this pattern of results suggests that the next candidate for consolidation would be semantics, and that differences between groups would occur higher up on the chain later in the instructional cycle (e.g., at the schematic and perhaps design-strategic levels). This expectation was confirmed: At the end of instruction in the second grade, there were no differences between instructional contexts for measures of syntax or semantics, suggesting consolidation. However, the schematic knowledge of students in the reflective context was greater than that of their counterparts in the inquiry context, and these between-group differences were even more pro- nounced at the strategic level (program design and debugging).

The progression of between-group differences in the fifth-grade lends further support to the joint effects of' reflection and the chain of consequences. After 1 month of instruction, there were no between-group differences at the lowest level of knowledge (syntax), but there were differences higher up in the chain (semantics, design strategic). The largest difference occurred at the highest level of knowledge. Assuming consolidation of syntax, the next candidates for consolidation would be the semantic and schematic 1evels:As expected, at the end of instruction, there were no differences between groups at either the semantic or schematic levels, but there were large differences at the design-strategic level.

Differences Between Grades

Reflective instruction was effective for both second- and fifth-grade students; the median effect size associated with statistically significant between-group differ-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


ences was .79 in the second grade and .93 in the fifth grade. However, we observed differences in the ways second and fifth graders made use of the mental tools provided to assist reflection. The daily summaries of the second graders were much shorter than those of their counterparts in the fifth grade, although there was a fair amount of evidence in the summaries students wrote at both grade levels of higher level identification (Vallacher & Wegner, 1985) of programming practices, such as "I broke the problem down today into procedures." Second-grade students made use of the programming templates when asked to do so by their instructor but rarely did so otherwise. However, fifth-grade students often made use of the programming templates to assist themselves in program design. In sum, in the second grade, programming templates served as concrete, inspectable objects to sustain instructor-student dialog, but, in the fifth grade, the templates served as worked examples for autonomous program design.

Another difference between grades was that second-grade students took longer to master programming skills, perhaps due to less availability of cognitive resources like working memory (Just & Carpenter, 1992). This conjecture was supported by largedifferences between grades on the measure of Logo working memory. The pattern ofresults evident at the end of instruction suggest that the ultimate benefit ofre- flective instruction for the older children was in program design (e.g., strategic knowledge), and for the younger children, the benefit was in both the overall organization of knowledge (schematic knowledge) and in program design.

Transfer

The results suggest that children in the reflective context at both grade levels transferred some components of learning Logo to related contexts. Transfer of specific skills like detecting errors is feasible to the extent to which acquisition and transfer tasks share mental units in common (Singley & Anderson, 1989). The results obtained generally support this line of work: Children in the reflective context received more effective instruction about skills like summarization and program debugging, and they transferred these skills more readily than did their counterparts. Also, the basis of the transfer was not mysterious; for example, measures of debugging in Logo were correlated highly with measures of detecting errors in text. At the same time, there were no differences with respect to either instructional context for the transfer of general cognitive skills such as detecting spatial or numeric patterns. The sole exception to this generalization was the finding in the second grade of increased sensitivity to the relation between old and new information for those participating in the reflective context, although, in some ways, this skill was explicitly targeted for instruction in the reflective context-by making debugging strategies more explicit, one attends more to the relation between previous and current states of a computer program (program bugs often arise from adding new features to old programs). Transfer of general skills was often the lodestone of earlier research

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

about the utility of learning programming languages, but more than a decade of research suggests otherwise. Perhaps the rationale can be summarized by analogy to a simple tool: Hammers don't make good carpenters, and Logo doesn't cultivate good habits of mind. However, both hammers and Logo can be useful tools, de- pending on particular goals and problems.

Incubating Habits of Mind

As noted earlier, one of the benefits of reflection for early advocates like Dewey was in the development of good habits of mind. In this instance, these habits refer to the development of principles about programming, design, and their interaction. In this research, students in the reflective context first took the perspective of the designers of Logo to learn the language and then designed products for a real audience. It is difficult to convey the excitement engendered by this design stance as students invented, for example, their own ways to turn and move the turtle or the in- sight gained as students compared their notations to those developed by the designers of Logo. Stated another way, student designers saw how invention and convention were related: For Logo primitives, like move and turn commands, they could relate their inventions for something like a turn (they invented ideas like turn right, slide left, etc.) to the conventions forced on them by the designers of Logo (i.e., RIGHT to turn right). We helped the second-grade students bridge this gap by programming Logo to accept their consensus about a sensible syntax and semantics for turns (TR ?h meant turn your body one half of a revolution). At a more complex level, students internalized programming practices, such as modularization, as beliefs because they had the opportunity to see their utility in practice. For example, in the fifth grade, students talked about the need to provide more explicit feedback about errors in their tutorial programs after watching other students try to use their programs. This growing awareness of audience was often apparent in their daily summaries; as one fifth-grade student noted, "Today I learned that the fourth-graders could not understand why they got the multiplication problem wrong! I'm going to have to give them some help." Recognition of the need for greater interactivity in programs grew into a design standard in the class (a convention) that was realized in different forms by different designers (inventions).

In summary, the results of this study are best interpreted by simultaneous consideration of the social and the cognitive. At the social level, reflective thinking was promoted and supported by a classroom of designers-not a mere collection of people, but an orchestrated repertoire of socially oriented practices, including teacher mentoring, a sense of audience, and a historic relation between convention and invention. Therefore, reflection was a form of mediated activity, sustained by an ensemble of tools and nurtured by specific forms of teaching (and peer) assis- tance. At the individual level, reflective thought resulted in more rapid acquisition and transfer of programming knowledge, performances tightly coupled with stu-

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

dents' internalization of socially constituted practices. Together, the joint plane of the social and the cognitive suggests that "cognitive change must be regarded as both a social and an individual process" (Newman, Griffin, & Cole, 1989, p. I), bringing us full circle back to Dewey and Mead.

ACKNOWLEDGMENTS

We thank Leona Schauble, Xiaodong Lin, Jim Kaput, and two anonymous reviewers for their thoughtful comments on an earlier version of this article. Thanks also to the teachers, children, and parents of thesugar CreekElementary School, Verona, WI.

REFERENCES

Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard University Press Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Brown, A. L. (1992). Design experiments: Theoretical and methodological challenges in creating com-

plex interventions in classroom settings. The Journal of the Learning Sciences, 2, 141-178. Brown, A. L., Bransford, J. D., Ferrara, R. A., & Campione, J. C. (1983). Learning, remembering, and

understanding. In P. H. Mussen (Ed.). Handbook of childpsychology: Vol. 3. Cognitive development (pp. 77-166). New York: Wiley.

Carver, S. M., & Klahr, D. (1986). Assessing children's LOGO debugging skills with a formal model. Journal of Educational Computing Research, 2(4), 487-525.

Case, R. (1978). Intellectual development from birth toadulthood: A Neo-Piagetian interpretation. In R. S. Siegler (Ed.), Children's thinking: What develops? (pp. 37-7 1). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Case, R. (1985). Intellectual development: Birth to adulthood. Orlando, FL: Academic. Clements, D. H. (1986). Effects of LOGC) andCAI environments on cognition and creativity. Journal of

Educational Psychology, 78.309-3 1.8. Clements, D. H. (1 990). Metacomponential development in a Logo programming environment. Journal

of Educational Psychology, 82, 141--149. Clements, D. H., & Merriman, S. (1988). Componential developments in LOGO programmingenviron-

ments. In R. E. Mayer(Ed.), Teaching andlearning computerprogramming: Multiple researchper- spectives (pp. 13-54). Hillsdale. NJ: Lawrence Erlbaum Associates, Inc.

Collins, A., Brown, J. S., & Newman, E. E. (1989). Cognitive apprenticeship: Teaching the crafts of reading, writing, and mathematics. In L. 9. Resnick (Ed.), Knowing, learning, and instruction: Es- says in honor of Robert Glaser. (pp. 453-494). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Cope, P., & Simmons, M. (1994). Some effects of limited feedback on performance and problem-solving strategy in a Logo microworld. Journal of Educational Psychology, 86,368-379.

Daneman, M., & Carpenter, P. A. (1983). Individual differences in integrating information between and within sentences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 561-581.

Day, J. D. (1986). Teaching summarization skills. Cognition and Instruction, 3, 193-210. DeCorte, E., Verschaffel, L., & Schrooten, H. (1992). Cognitive effects of learning to program in Logo:

A one-year study with sixth graders. In E. DeCorte, M. C. Linn, H. Mandl, & L. Verschaffel (Eds.), Computer-based learning environments and problem solving (pp. 207-228). New York: Springer-Verlag.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


Detterman, D. K., & Stemberg, R. J. (1993). Transfer on trial: Intelligence, cognition, and instruction. Norwood, NJ: Ablex.

Dewey, J. (1933). How we think. A restatementof the relation of reflective thinking to the educativepro- cess. Lexington, MA: Heath.

Fay, A. L., & Mayer, R. E. (1987). Children's naive conceptions and confusions about LOGO graphics commands. Journal of Educationul Psychology, 79, 254-268.

Fay, A. L., & Mayer, R. E. (1988). Learning Logo: A cognitive analysis. In R. E. Mayer (Ed.), Teaching and learning computerprogrammrng: Multiple researcher perspectives (pp. 55-74). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Flavell, J. H. (1985). Cognitive development. Englewood Cliffs, NJ: Prentice-Hall. Flavell, J. H., Green, J. L., & Flavell, E R. (1985). The road not taken: Understanding the implications of

initial uncertainty in evaluating spatial directions. Developmental Psychology, 21,207-216. Geva. E., & Cohen, R. (1987, April). Predicting children's success in Logoprogramming. Paper pre-

sented at the annual meeting of the American Education Research Association, Washington, DC. Harel, 1. (199 1). Children designers: Interdisciplinary constructionsfor learning and knowing mathe-

matics in a computer-rich school. Norwood, NJ: Ablex. Just, M. A,, & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in

working memory. Psychological Review, 99, 122-149. Klahr, D., & Carver, S. M. (1988). Cognitive objectives in a LOGO debugging cuniculum: Instruction,

learning, and transfer. Cognitive I'sychulogy, 20.362404. Lawson, B. (1990). How designers think (2nd ed.). Boston: Butterworth Architecture. Lehrer, R., Guckenberg, T., & Lee, 0. (1988). Comparative study of the cognitive consequences of in-

quiry-based Logo instruction. Journal cf Educational Psychology, 80,543-553. Lehrer, R., Guckenberg, T., & Sancilic~, L. (1988). Influences of LOGO on children's intellectual devel-

opment. In R. E. Mayer (Ed.), Tetrching and learning computer progmmming: Multiple reseurch perspectives (pp. 75-1 10). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Lehrer, R., & Littlefield, J. (1991). Misconceptions and errors in Logo: The role of instruction. Joumul of Educational Psychology, 83, 124-133.

Lehrer, R., & Littlefield, J. (1993). Relationships among cognitive components in Logo learning and transfer. Journal of Educational Psychology, 85,3 17-330.

Lehrer, R., Littlefield, J., Wottreng, B., & Youngeman, N. (1993). Seeding mindstorm: Using Logo in the elementary classroom. Fontana, WI: Interactive Education Technologies.

Linn. M. C. (1985). The cognitive consequences of programming instruction in classrooms. Educu- tional Researcher, 14, 14-16.25-29.

Littlefield, J., Delclos, V., Bransford, .I., Clayton, K., &Franks, J. (1989). Some prerequisites for teaching thinking: Methodological issues in the study of LOGO programming. Cognition and Instruc- tion, 6 , 331-366.

Littlefield, J., Delclos, V., Lever, S.. Clayton, K., Bransford, J., &Franks, J. (1988). Learning LOGO: Method of teaching, transfer of general skills, and attitudes toward school and computers. In R. E. Mayer (Ed.), Teaching and learning computerprogmmming: Multiple research perspectives (pp. 11 1-135). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Maxwell, S. W., Delaney, H. D., & Dill. C. A. (1984). Another lookat ANCOVA versus blocking. Psy- chological Bulletin, 93, 13& 147.

Mayer, R. E. (1985). Learning in complex domains: A cognitive analysis of computer programming. In G. Bower (Ed.), Thepsychology oj'learning andmotivation (Vol. 19, pp. 89-130). Orlando, FL: Ac- ademic.

Mayer, R. E. (1992). Teaching for transfer of problem-solving skills to computer programming. In E. DeCorte, M. C. Linn, H. Mandl, & L. Verschaffel (Eds.), Computer-bused learning environments and problem solving (pp. 193-206). New York: Springer-Verlag.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


McKeithen, K. B., Reitman, J. S., Rueter, H. R., & Hirtle, S. C. (1981). Knowledge organization and skill differences in computer programmers. Cognitive Psychology, 13, 307-325.

Mead, G. H. (1910). Social consciousness and the consciousness of meaning. Psychological Bulletin, 7, 397-405.

Mead, G. H. (1932). The philosophy of the present. Chicago: University of Chicago Press. Naveh-Benjamin, M., McKeachie, W. J., Lin, Y .-G., &Tucker, D. (1986). Infemng students' cognitive

structures and their development using the "ordered tree technique." Journal of Educational Psy- chology, 78,279-288.

Newell, A,, & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice-Hall. Newman, D., Griffin, P.. & Cole, M. (1989). The construction zone: Working for cr~gnirive churzge in the

school. Cambridge, England: Cambridge University Press. Olson, D. R., & Astington, J. W. (1993). Thinking about thinking: Learning how to take statements and

hold beliefs. Educational Psychologist, 28, 7-23. Palinscar, A. S., & Brown, A. L. (1984). Reciprocal teaching of comprehension-fostering and monitor-

ing activities. Cognition and Instruction, 1 , 117-175. Powers, W. T. (1973). Behavior: The control ofperception. Chicago: Aldine. Pressley, M. (1986). The relevance of the good strategy user model to the teaching of mathematics. Edu-

cationul Psychologist, 21, 139-161. Pressley, M., Goodchild, F., Fleet, J., Zajchowski, R., & Evans, E. D. (1989). The challenges of class-

room strategy instruction. Elementary School Journal, 89,301-342. Reitman, J . S., & Rueter, H. H. (1980). Organization revealed by recall orders and confirmed by pauses.

Cognitive Psychology, 12, 554-581. Salomon, G., & Perkins. D. N. (1989). Rocky roads to transfer: Rethinking mechanisms of a neglected

phenomenon. Educationul Psychologist, 24, 1 13-142. SchGn, D. A. (1990). Educuting the reflective practitioner. San Francisco: Jossey-Bass. Siegler, R. S. (1991). Children's thinking. Englewood Cliffs. NJ: Prentice-Hall. Simon, H. A. (1969). The sciences of the artificial. Cambridge, MA: MIT Press. Singley, M. K., & Anderson, J. R. (1989). The rransfer rfcognitive skill. Cambridge, MA: Harvard Uni-

versity Press. Soloway, E., Spohrer, I. , & Littman, D. (1988). E Unum Pluribus: Generating alternative designs. In R.

E. Mayer (Ed.), Teaching and learning computer programming: Multiple research perspectives (pp. 137-152). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Stanley, J. C. (1971). Reliability. In R. 1,. Thorndike (Ed.), Educational measurement(pp. 356442). Washington, DC: American Council on Education.

Swan, K. (1989). Logo programming and the teaching and learning of problem solving. Journul ofArti- ficial Intelligence in Education, 1, 73-92.

Symons, S., Synder, B. L., Cariglia-Bull, T., & Pressley, M. (1989). Why be optimistic about cognitive strategy instruction? In C. B. McCormick, G. E. Miller, & M. Pressley (Eds), Cognitive strategy research: From basic research to educcltionalapplications. (pp. 3-32). New York: Springer-Verlag.

Vallacher, R. R., & Wegner, D. M. (1985). A theory of action identification. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. D

ownl

oade

d by

[U

nive

rsity

of

Cal

ifor

nia,

San

Fra

ncis

co]

at 2

3:00

16

Dec

embe

r 20

14

APPENDIX A Sample Lesson: Procedures With REPEAT

Differences between the inquiry and reflective contexts (reflective instruction) are indicated by italics, with comments about the functions of different components of the reflective context displayed in brackets.

Lesson Overview

The syntax and semantics of the REPEAT control structure are introduced. Anal- ogy is made between REPEAT and other everyday activities, including breathing. A "baseball" game helps children reflect on the semantics of REPEAT.

Establishing the Need for REPEAT

"Sometimes in Logo we do the same thing over and over again. For example, who can remember how we made the square? Let's look at that. What do you notice?' (The discussion usually leads to some expression of having the same commands expressed four times.)

"lf you were inventing Logo and you wanted to do something over and over again without typing every command every time, what command would you make up? How would it work?" Compare and contrast children's solutions. Compare them to Logo's REPEAT and ask children what they think Logo's inventors might have had in mind. Ask children to evaluate the choice made by the inventors-is it good? bad? and why. [This relates to the design stance component.]

Bridge to Breathing

"What do you do over and over again everyday?'(Elicit children's ideas.) "How about breathing? You breathe in and out, right? So that's like using REPEAT."

"Let's try this with turtle drivers. When the turtle driver says so, the turtle will breathe IN. When the turtle driver says so, the turtle will breathe OUT. Now let's try it, What do you notice? It's a little trouble to keep having to say IN and OUT is- n't it? With REPEAT we could just say something like this: REPEAT 1000 [IN OUT] and that will keep the turtle green instead of blue!" [Turtle drivers are a way to objectify the semantics of the command "out there" in the action of breathing. The turtle drivers provide additional emphasis on bridging between the anchor of breathing and the syntadsemantics of the command.]

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


Syntax and Semantics of REPEAT

"Here's how REPEAT works. First we say REPEAT, then we give the turtle a list of instructions that we want it to REPEAT. The list of instructions is one of the inputs to the REPEAT procedure. Does anyone know what a list is? When you go shopping, what do you use? In Logo, we make a list of things to do by using the square brackets on the keyboard, like this: [things to do], like [FD 50 R 901. When else did we use lists? (Tie back to use of LABEL command.)"

"So when we breathe we might say REPEAT 1000 [IN OUT] (IN for breathe in and OUT for breathe out). Let's look again at the square. How could we use REPEAT to make a square? What would you put in the list? (Things to do are put into the list just like the grocery list when we write things to get at the store.)

REPEAT how many times [things to do] REPEAT how many times [FII 50 RT 901 REPEAT 4 [ FD 50 RT 901 The turtle works like this. First it reads REPEAT so it knows that it is going to

have to do something over and over again. Then the turtle looks for how many times it will have to do something over and over again. Then it looks inside the list for what it is supposed to do. So for a square it says to itself: Oh, I should REPEAT 4 times FD 50 RT 90.

"Get the Baseball page. Type Playball. You will see a baseball diamond and a player at home plate. If you type, REPEAT 3 [RUNBASES], what happens? (the turtle runs to third base). So we can move the turtle around the bases with REPEAT number-of-times [RUNBASES]. RB can be used instead of RUNBASES. HP will always move the turtle to home plate."

"Starting at home, which base will the turtle land on ifyou type REPEAT 3 [RBI? How about REPEAT 7 (RBI? How about REPEAT 11 [RBI? What's the pattern? What about REPEAT 16 [RBI?" [ Use of the baseball game lets students reflect about the general syntax and structure of REPEAT outside of the immediate context of drawing a graphic. The microworld helps make the action of REPEAT more inspectable.]

Entrapment

For the reflective group, use the context of turtle drivers to predict what will happen for each of the three challenges that are listed here. For the inquiry group, ask children to predict for themselves what will happen. For both groups, the key challenge is the third, as many children believe that the turtle wiiI first go FORWARD 20 steps and then turn 180 degrees.

( I ) REPEAT 2 [BK 31

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

(2) REPEAT 3 [FD 51 (3) REPEAT 2 [FD 10 RT 901

Practice and Extension

The context of turtle drivers is used in the reflective group, only. The inquiry group uses the computer.

(1) How many different ways can you use REPEAT to get the turtle to turn RT 90? LT 180? FD 1 O? [FD 10 RT 90]?

(2) Someone made this with REPEAT, TO SPINSQUARE REPEAT 12 [ SQUARE RT 301 END What is going on here? How can such a complex pattern come from such a sim-

ple thing? (3) REPEAT 2 [RT ? ] = RT 90 (4) REPEAT 10 [FD 10 BK 10 RT 361 (5) REPEAT 3 [BK 101 = REPEAT ? [BK 5 ] (6) Will REPEAT 2 [FD 10 RT 901 make the same path as FD 20 RT 180? (7) REPEAT 2 [REPEAT 2 [FD 30 RT 9011 (Fifth grade, only) For the reflective group, children write a short summary of what they learned

about REPEAT. Also, the salient points of REPEATare available as aprogramming template, similar in style to that described next for modularization. [The summaries and template are the other two components of the reflective instruction.]

DESCRIPTION OF A PROGRAMMING TEMPLATE ON MODULARIZATION

(Numbers refer to successive screen displays) 1. BREAKING a problem DOWN means: Finding the parts and Writing PROCEDURES for each part TO GO.TO.SCHOOL Wake.up Wash.face Brush.teeth Comb.hair Eat.breakfast Walk END

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


2. The house has three parts HOUSE = Bottom + Roof + Penmove

FIRST, you need the bottom part

SQUARE can make the bottom part

TO SQUARE F 5 R 9 0 F 5 R 9 0 F 5 R 9 0 F 5 R 9 0 END

3. NEXT, we can draw the ROOF TRIANGLE can make the roof TO TRIANGLE F 5 R 120 F 5 R 120 F 5 R 120 END

4. This is what happens if you just put square and triangle together TO HOUSE SQUARE TRIANGLE END

5. So we need to move the pen after we make the square TO MOVEPEN P U F 5 L 9 0 B 5 P D END

6. Now, we put them all together TO HOUSE SQUARE MOVEPEN TRIANGLE END

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


So if you enter HOUSE at the command center

You have taught the Turtle How to draw a house (Other interactive examples follow.)

APPENDIX B Samples of Children's Programming

Grade 2 Movie Program

TO movie castle story END

TO castle castle 1 castle2 castle3 castle4 castle5 shadec END

TO story PRINT [Once upon a Time there lived two girls in a castle] PRINT [They were very happy] END

TO castle l SETSH 0 P U L 9 0 F 10L 90

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


P D F 5 L 9 0 F 1 2 L 9 0 F 5 END

TO castle2 L 9 0 F l L 9 0 F l R 9 0 F I R 9 0 F 1 END

TO castle3 L 9 0 F l L 9 0 F 3 R 9 0 F 2 R 9 0 END

TO castle4 F 3 L 9 0 F 1 L90 REPEAT 2 [F 1 R 901 F l L 9 0 END

TO castle5 F I L 9 0 F 3 R 9 0 F 2 R 9 0 F 3 L 9 0 F 2 END

TO shadec PU L 145 F 2 PD SETSH 29 SHADE END

Grade 5 Madlib Program

TO Madlib Mad CT S torys END

This does the MadLib. TO Mad M a d l i END

Madli asks the questions. TO Storys story. 1 story.2 story.3

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


story .4 story.5 END

This makes it easier to look at the story (for Madlib). TO m PR [Give me the name of a girl.] MAKE "Girl RW END

TO a PR [Give me a sickness.] MAKE "Sickness RW END

TO d PR [Give me a bodypart.] MAKE "Bodypart RW END

TO 1 PR [Give me an action.] Make "Action RW END

TO i PR [Name a famous person.] MAKE "person READLIST END

TO Story. 1 (PR [Please excuse] :girl) END

TO Story.2 (PR [She has the] :sickness) END

TO Story.3 (PR [and her] :bodypart) END

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014


TO Story.4 (PR [is] :action) END

TO Story.5 (PR [off. Signed, ] :person) END

Dow

nloa

ded

by [

Uni

vers

ity o

f C

alif

orni

a, S

an F

ranc

isco

] at

23:

00 1

6 D

ecem

ber

2014

Documents

Reflective Teaching of Logo