This article was downloaded by: [York University Libraries]On: 22 November 2014, At: 02:00Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office:Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Behaviour & Information TechnologyPublication details, including instructions for authors and subscriptioninformation:http://www.tandfonline.com/loi/tbit20

Persuasiveness of expert systemsJaap J. Dijkstra a , Wim B. G. Liebrand b & Ellen Timminga ba Social Science Information Technology , University of Groningen , GroteRozenstraat 15, Groningen, 9712 TG, The Netherlands E-mail:b Social Science Information Technology , University of Groningen , GroteRozenstraat 15, Groningen, 9712 TG, The NetherlandsPublished online: 08 Nov 2010.

To cite this article: Jaap J. Dijkstra , Wim B. G. Liebrand & Ellen Timminga (1998) Persuasiveness of expertsystems, Behaviour & Information Technology, 17:3, 155-163, DOI: 10.1080/014492998119526

To link to this article: http://dx.doi.org/10.1080/014492998119526

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)contained in the publications on our platform. However, Taylor & Francis, our agents, and ourlicensors make no representations or warranties whatsoever as to the accuracy, completeness, orsuitability for any purpose of the Content. Any opinions and views expressed in this publication arethe opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis.The accuracy of the Content should not be relied upon and should be independently verified withprimary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoevercaused arising directly or indirectly in connection with, in relation to or arising out of the use of theContent.

This article may be used for research, teaching, and private study purposes. Any substantialor systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, ordistribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use canbe found at http://www.tandfonline.com/page/terms-and-conditions

Persuasiveness of expert systems

JAAP J. DIJKSTRA, WIM B. G. LIEBRAND and ELLEN TIMMINGA

Social Science Information Technology, University of Groningen, Grote Rozenstraat 15, 9712 TG Groningen,

The Netherlands; email j.j.dijkstra@ ppsw.rug.nl

Abstract. Expert system advice is not always evaluated by

examining its contents. Users can be persuaded by expertsystem advice because they have certain beliefs about advice

given by a computer. The experiment in this paper shows that

subjects (n = 84) thought that, given the same argumentation,expert systems are more objective and rational than humanadvisers. Furthermore, subjects thought a problem was easier

when advice on it was said to be given by an expert system while

the advice was shown in production rule style. Such beliefs canin¯ uence expert system use.

1. Introduction

Do users judge advice of expert systems by examining

its contents? This paper argues that users’ beliefs about

expert systems in¯ uences their evaluation of the advice.

It is important to know how users assess advice of expert

systems, because they have to neglect the advice when

the expert system is incorrect (Waldrop 1987, Hollnagel

1989, Chen 1992, Suh and Suh 1993, Berg 1995).

Therefore, modern expert systems are equipped with

explanation functions that give users the opportunity to

examine the inference (Berry and Broadbent 1987,

Waldrop 1987, Wognum 1990, Bridges 1995, Ye and

Johnson 1995). But do users consult these explanation

functions? Are users interested in the argumentation

behind the advice? Theory on persuasion shows that

advice is often judged without an examination of the

arguments. Beliefs held by a person about the adviser

can in¯ uence his or her attitudes towards the advice

(Petty and Cacioppo 1986a, b). Therefore, it is possible

that user beliefs and attitudes in¯ uence the use of expert

system advice.

In M anagement Information System research the

topic of user beliefs and attitudes has been studied

extensively (Robey 1979, Ginzberg 1981, Baroudi et al.

1986, Barki and Hartwick 1989, 1994, Igbaria 1993,

Hartwick and Barki 1994, Pare and Elam 1995,).

Especially user satisfaction and perceived usefulness,

are believed to be important factors for system use (Ives

et al. 1983, Davis 1989, 1993, Davis et al. 1989, Galletta

and Lederer 1989, Igbaria and Nachman 1990, Gatain

1994, Igbaria et al. 1994, Iivari and Ervasti 1994, Keil et

al. 1995). But, user satisfaction will not always indicate

whether computer support is desirable (Srinivasan 1985,

M elone 1990, Szajna 1993). Some studies underline that

users can be wrong in their judgement on the usefulness

of intelligent computer support. Kottemann et al. (1994)

show that redundant what-if help increases users’

con® dence and Davis et al. (1994) point out that

irrelevant information in an information system weak-

ens performance but increases the users’ con® dence.

Similar results were reported by Aldag and Power (1986)

and Will (1992). Cornford et al. (1994) warn for the

in¯ uence an imperfect medical expert system might have

on medical professionals in developing countries.

Dijkstra (1995) found that the incompleteness and

incorrectness of argumentations made by a legal expert

system were hardly noticed by the lawyers who used it.

These studies show that users sometimes make

judgements on expert system advice that are not

supported by the contents of the argumentation. User

acceptance of expert system advice is not always based

on a thorough examination of the advice. This notion is

supported by the Elaboration Likelihood Model (Petty

and Cacioppo 1986a, b), a theory on persuasion that

states that people are likely to make their judgement on

peripheral cues if they are less motivated or unable to

judge a message on its contents. Persuasiveness of the

source of the message is an important peripheral cue.

For example, a doctor’ s advice is believed to be

important because it is made by a doctor, not because

of a positive evaluation of the message by the patient.

According to the Elaboration Likelihood M odel, user

acceptance of expert system advice without thorough

examination of correctness or applicability might be

explained by persuasiveness of the expert system.

The motivation to examine the contents of a message

is triggered by personal relevance, responsibility, and

need for cognition. Need for cognition indicates whether

a person likes to examine an argument (Cacioppo and

BEHAVIOUR & INFORMATION TECHNOLOGY, 1998, VOL. 17, NO. 3, 155 ± 163

0144-929X/98 $12.00 Ó 1998 Taylor & Francis Ltd.

Dow

nloa

ded

by [

Yor

k U

nive

rsity

Lib

rari

es]

at 0

2:00

22

Nov

embe

r 20

14

Petty 1982, Priester and Petty 1995). A person who does

not like to study an argument is more likely to make a

judgement on peripheral cues. Todd and Benbasat

(1992, 1994a, b) found that imperfect use of decision

support systems occurs more often when users want to

reduce their cognitive eŒort. Thus, need for cognition

can be an important indicator of how a person will use

an expert system.

The research presented in this paper is on persuasive-

ness of expert systems. It tries to explain why users of

expert systems sometimes put their trust into redundant

or even incorrect information presented by an expert

system. In line with the Elaboration Likelihood Model,

it is thought that users have certain beliefs about advice

given by an expert system that, regardless of its contents,

makes them agree easily. This research tries to identify

some of these beliefs that people might have about

advice given by an expert system.

In the experiment, subjects had to judge advice on

several cases. Some subjects were told that the arguments

were made by expert systems and some were told they

were made by humans. In the expert system group, half

of the subjects got arguments in natural language (the

same as in the human group), the other half got them in

production rule format (see table 1). Subjects were asked

to ® ll out a questionnaire on what they thought of the

arguments and conclusions. If subjects base their

judgement only on the contents of the advice, this

evaluation will not diŒer when the adviser is said to be a

human or an expert system. However, it is hypothesized

that an expert system is a persuasive message source in

comparison with human advisers. Research has shown

that the advice of an expert system can be convincing

even when its arguments are weak (Kottemann and

Remus 1987, Dijkstra 1995). Thus, it is expected that the

expert system groups will score higher on the persua-

siveness factors (H1). Furthermore, arguments presented

in a production rule format are more di� cult to judge

because people are not used to reading production rules.

Therefore, it is hypothesized that subjects who got the

argument of the expert system in a production rule

format will score higher on the persuasiveness factors

than subjects who got the argument in natural language

(H2). In addition, it is expected that subjects with a low

need for cognition are less motivated to study the

argument and, consequently, are more likely persuaded

by an expert system. Thus, subjects with a low need for

cognition will score higher on the persuasiveness factors

in the expert system conditions than subjects with a high

need for cognition (H3).

2. Method

2.1. Subjects

Students were asked to participate in the experiment

through an advertisement in a local university news-

paper. They were paid 12 Dutch guilders (8 US dollars).

Eighty-® ve students of several faculties participated.

After randomly assigning the subjects to the conditions,

28 subjects were told that the arguments were made by a

human, 28 subjects were told that the arguments were

made by an expert system, and 29 subjects received the

same arguments in a production rule style and were told

that they were made by an expert system.

2.2. Apparatus and procedure

Table 2 gives an overview of the ® ve phases of the

procedure. The procedure started with a pretest on

subjects’ experience with computers and the knowledge

domains involved (Phase 1). Subjects’ need for cognition

was tested with a questionnaire developed by Cacioppo

and Petty (1982) and translated in Dutch by Pieters et al.

(1987). After the pretests, subjects carried out a simple

number-circling task (Phase 2). Subjects were asked to

circle every 1, 5, and 7 in a table of 875 random

numbers. This test was used as a distraction to reduce

the eŒect of the pretest. The test was also used as a check

on the need for cognition test (see Phase 4).

In the main part of the test (Phase 3), four cases with a

conclusion and argument were handed out. In each case,

a person’ s problem was presented. The problem was

described, with an advice on it said to be given by either

an expert system or a human (for an example see the

Appendix). The cases were on dyslexia, cardiology, law,

and train tickets. After each case, the subject ® lled out a

questionnaire on the conclusion and argument. All

subjects evaluated, in random order, the same cases with

the same argument and conclusions. The only diŒerence

between Conditions 1 and 2 was the message source

mentioned (a human in Condition 1 and an expert

system in Condition 2). In Condition 3 the same

argumentation was presented as in Conditions 1 and 2

but in a production rule style and with an expert system

as message source mentioned (see table 1 and the

Appendix).

J. J. Dijkstra et al.156

Table 1. The experimental groups.

Expert system adviser

Human adviserNaturallanguage Production rules

Condition 1 2 3

Dow

nloa

ded

by [

Yor

k U

nive

rsity

Lib

rari

es]

at 0

2:00

22

Nov

embe

r 20

14

The items used in the questionnaire were based on

research on interpersonal persuasiveness and on qualities

people expect from expert systems. Important interper-

sonal persuasiveness factors are credibility (competence

and trustworthiness), liking, similarity, and consensus

(Petty and Cacioppo 1986b, Worchel et al. 1988, O’Keefe

1990). Qualities people associate with expert systems are

objectivity and rationality (Boden 1990, Berg 1995).

Items were designed for all these constructs, with two

items on the di� culty of the case, making a questionnaire

of 23 items (see table 3 for item wordings). A factor

analysis was used to distinguish factors thought to be

important for judging expert system advice.

At the end, subjects carried out a complex number-

circling task (Phase 4). They had to circle every 3, every

6 when a 7 followed, and every second 4 in a table of 600

numbers. It was expected that subjects with a high need

for cognition liked the di� cult number-circling task

better than the simple one. The ® nal questionnaire

(Phase 5) was a recall test on the information presented

in the dyslexia, cardiology, law, and train ticket cases.

2.3. Construction of the need for cognition scale

For the construction of the need for cognition scale, 13

of the 18 items from a test developed by Cacioppo et al.

(1984) were used. Pieters et al. (1987) dropped Items 5,

13, and 16 after translating and testing the scale in Dutch

(for item wording see Cacioppo et al. 1984). Item 18 was

dropped because, in a small pilot study, subjects thought

it was incomprehensible. Of the 14 items used in the test,

Item 9 was dropped because it correlated very low with

the other items (the corrected item total correlation was

0.11). Cronbach’ s alpha was 0.79 for the 13 items used.

As suggested by Cacioppo and Petty (1982), the need for

a cognition scale was constructed by collapsing the item

scores. Three 33% percentiles were used to make groups

with a high, average, and low need for cognition.

With two 6-point Likert scale items subjects were

asked whether they liked the di� cult number-circling

task better than the simple one (Phase 4, see table 2).

Cacioppo and Petty (1982) used a similar test to verify

the need for a cognition scale. As predicted, subjects

with a high need for cognition (M = 8.1) liked the

di� cult task better than subjects with a low need for

cognition (M = 6.8), t(49) = 2.44, p < 0.01. Also, scores

on the recall questions (Phase 5, see table 2) were used

to test the need for cognition scale. According to the

Elaboration Likelihood Model, a high need for

cognition would invoke a thorough examination of

the cases. Therefore, subjects with a high need for

cognition were expected to score better on recall in

comparison with subjects with a low need for cogni-

tion. The recall scale was constructed by taking the

number of correct answers on eight good/wrong recall

questions. As predicted, subjects with a high need for

cognition (M = 6.1) scored better on recall than

subjects with a low need for cognition (M = 5.5),

t(49) = 2.02, P < 0.05.

2.4. Construction of the computer experience scale

A scale for computer experience was constructed by

using four questions from the pretest (Phase 1, see table

2). The ® rst question was on frequency of computer use

(never, monthly, several times a week, weekly, daily;

scoring 0 ± 4). The second question was on diversity of

computer use (`Do you use a computer for: word-

processing, databases, electronic mail, programming,

games’ ; scoring the number of applications: 0 ± 5). The

other two questions were 6-point Likert scale items on

computer expertise and expert system knowledge. A

Principal Component Analysis (PCA) showed one

factor, accounting for 65% of the variance. Cronbach’ s

alpha was 0.81. The PCA factor scores were taken as an

indication of computer experience. Although not

hypothesized, a signi® cant correlation was found

between need for cognition and computer experience,

r = 0.36, P < 0.001; subjects with a high need for

cognition had more computer experience.

Persuasiveness of expert systems 157

Table 2. An overview of the procedure.

Phase 1: Prestests Phase 2: Distraction Phase 3: Main test Phase 4: Distraction Phases 5: Recall

Sex, age, study (3 items)Knowledge on: dyslexia,

cardiology, train

tickets, law (4 items)Computer experience

(4 items)Need for cognition

(14 items)

Simple number ±circling test

Four cases in randomorder with after each

case a questionnaire

of 23 items

Di� cult number ±circling test and a

short questionnaire

on the number ±circling test

Eight recall questionson the cases

Dow

nloa

ded

by [

Yor

k U

nive

rsity

Lib

rari

es]

at 0

2:00

22

Nov

embe

r 20

14

2.5. Construction of the persuasion scales

In the main part of the test, subjects had to judge 4

cases. After each case they ® lled out a questionnaire

(Phase 3, see table 2). Thus, subjects scored each item

four times. Simultaneous Component Analysis (SCA;

Kiers and Ten Berge 1989, Kiers 1990) with a varimax

rotation was used for summarizing the items by means

of a small number of components. In SCA, component

weights are found that de ® ne components that optimally

account for the variance in all four cases simultaneously.

These components have the same interpretation in all

four cases, because their weights are equal and apply to

the same items. The results are presented in table 3.

The SCA showed four factors: compliance, objectiv-

ity/rationality, easiness of the case, and authority. The

compliance and authority factors were based on the

interpersonal persuasiveness constructs. Items 2, 6, 10,

20, and 23 were used to measure authority. They

constitute the fourth factor. The other items based upon

interpersonal persuasiveness constructs constitute the

® rst factor: compliance. Its items are on trustworthiness

(Items 8, 11, and 21), liking (Items 9 and 22), similarity

(Item 19), and consensus (Items 3, 4, 5, 7, and 13).

Objectivity (Items 15, 16, and 18) and rationality (Items

1 and 17) make the second factor. Easiness of the case

(Items 12 and 14) is the third factor. Table 4 gives an

overview of the variance accounted for by the factors

found.

A Mokken scale analyses reliability test (Molenaar et

al. 1994) on the selected items per case supported the

constructed factors. Reliabilities found were about 0.95

for compliance, 0.65 for objectivity/rationality, 0.75 for

authority, and 0.40 for easiness of the case. In the

J. J. Dijkstra et al.158

Table 3. SCA factor weights for all cases and factor loadings for the law case (between brackets) of the items used in the main test.Absolute weights lower than 0.15 and absolute loadings lower than 0.35 are left out.

Item

number Item wording

Factor 1:

Compliance

Factor 2:Objectivity

rationality

Factor 3:

Easiness case

Factor 4:

Authority

3.

4.5.

7.

8.

9.*

11.

13.*19.21.

22.

I fully agree with the way the problem

is solved.

This way of problem solving is tempting.I wouldn’ t have any questions after

this explanation.

More often problems should be solved

this way.I have full con® dence in the way the

problem is solved.I wouldn’ t have tolerated such a treatment

if I had a similar problem.

The problem is solved reliably.

This conclusion is a bad one.I would have solved the case in a similar way.Inference and conclusion are trustworthy.

The problem is dealt with sympathetically.

0.34

0.260.23

0.24

0.33

0.33

0.34

0.290.310.20

0.24

(0.88)

(0.78)(0.68)

(0.84)

(0.91)

(0.73)

(0.82)

(0.74)(0.83)(0.61)

(0.56)

0.15

0.21

0.16

0.16

(0.53)

(0.50)

(0.57)(0.55)

(0.36)

1.

15.16.*

17.

18.

The problem is solved in a rational way.

The conclusion is objective.The problem is dealt with in an emotional

way.The problem is solved step by step.

The problem is dealt with unbiased.

0.33

0.450.51

0.26

0.48

(0.46)

(0.67)(0.56)

(0.57)

(0.70)

0.20

0.19 (0.51)

12.14.

This case is easy.A child could have come to this conclusion.

(0.42) 0.25 0.480.69

(0.55)(0.86)

Ð 0.17

2.

6.

10.20.

23.

The solution of the problem involves a lot

of thinking.The problem is analysed thoroughly.

The analyses shows great expertise.The problem is solved in a powerful way.

The solution is authoritative.

0.18 (0.60)

(0.68)(0.49)

Ð 0.32

Ð 0.210.18

( Ð 0.51) 0.32

0.19

0.320.41

0.60

(0.60)

(0.42)

(0.62)(0.69)

(0.73)

*reverse scoring is used on these items.

Dow

nloa

ded

by [

Yor

k U

nive

rsity

Lib

rari

es]

at 0

2:00

22

Nov

embe

r 20

14

dyslexia case reliabilities were a bit lower for objectivity/rationality and easiness of the case.

3. Results

All subjects ® nished their tasks in about one hour.

One subject did not ® ll out a questionnaire completely

and was left out of the analyses. No diŒerences could be

explained by sex, age, or study. For examining the

hypotheses, a 3 (condition) ´ 3 (need for cognition)

repeated measurement MANOVA was performed on

compliance, objectivity/rationality, easiness of the case,

and authority SCA scores, taking the four cases as

within factor. For the analysis, SCA factor scores were

used. Scores mentioned in ® gures 1 ± 4 are standardized

SCA factor scores. Results are reported below.

3.1. Compliance and authority

The SCA showed that compliance and authority were

highly correlated; the scores on the four cases showed

correlations between 0.51 and 0.66. An SCA can ® nd

constructs that correlate because several data sets (the

scores on the four cases) are factorized simultaneously.

Although compliance and authority are two separate

constructs (this was con® rmed in an analysis on the

indicators of the compliance and the authority construct),

they are related and therefore, they are discussed together.

Compliance and authority are both interpersonal

persuasiveness factors. W hen these factors would be

important for explaining persuasiveness of expert

systems, their scores should be signi® cantly higher in

the expert system conditions. However, there was only a

small within eŒect of the interaction between condition

and case diŒerence on compliance, F(6,225) = 2.22, P <0.05 (see ® gure 1). For the train ticket and law case the

compliance scores were a bit higher in the production

rule condition, but for the dyslexia and cardiology case

the scores were lower in the production rule condition

when compared with the expert system with natural

language condition. The dyslexia and cardiology case

were more technical and di� cult to understand than the

train ticket and law case were. Maybe this caused

negative feelings when production rules made the advice

even more technical. However, as ® gure 1 shows, the

eŒect is very small. On authority there was no signi® cant

between or within eŒect from the conditions.

Contrary to the expectation that subjects with a low

need for cognition would score higher for compliance and

authority in the expert system conditions, they scored

lower (see ® gure 2). Apparently, subjects with a low need

for cognition assigned less authority to the expert systems

than subjects with a high need for cognition. This goes

especially for the expert systems with a production rule

argumentation (t(12) = 2.85, P < 0.01).

The cases had a signi® cant within eŒect on both

factors (P < 0.001): compliance F(3,225) = 28.96;

authority F(3,225) = 5.51. Scores on compliance and

authority were much aŒected by the cases.

3.2. Objectivity/rationality

The objectivity/rationality factor seems to be an

expert system persuasiveness factor. There was a

signi® cant between eŒect of the conditions on the

scores, F(2,75) = 6.01, P < 0.01 (see ® gure 3). Subjects

thought the expert system to be more objective and

rational, especially when production rules were used.

There was a signi® cant within eŒect of the case

diŒerence on the objectivity/rationality scores,

Persuasiveness of expert systems 159

Figure 1. Compliance with the four cases.

Table 4. Percentage of variance accounted for by the SCA factors.

Case TotalFactor 1:

Compliance

Factor 2:

Objectivityrationality

Factor 3:Easiness case

Factor 4:Authority

DyslexiaLaw

CardiologyTrain ticket

59.958.7

61.364.9

40.936.3

37.743.1

8.610.8

16.518.6

7.17.6

8.57.2

25.317.8

22.326.8

Dow

nloa

ded

by [

Yor

k U

nive

rsity

Lib

rari

es]

at 0

2:00

22

Nov

embe

r 20

14

F(6,225) = 4.68, P < 0.01, thus, case diŒerence also

in¯ uenced objectivity/rationality scores.

Low need for cognition did not increase the eŒect of

the conditions. However, need for cognition seemed to

in¯ uence the evaluation. Figure 3 suggests that subjects

with a high need for cognition expected objective and

rational advice when an expert system was mentioned as

message source; subjects with a low need for cognition

increased their expectations of objectivity and ration-

ality when production rules were shown as well. The

diŒerence between high and low need for cognition was

signi® cant in the expert system with natural language

condition (t(19) = 2.17, P < 0.05).

3.3. Easiness of the cases

As expected, there was a signi® cant within eŒect of

case diŒerence on easiness of the case, F(3,225) = 36.70,

P < 0.01, but there was no interaction with other factors.

The train ticket case was judged to be easier than the

other cases. More interesting, however, was the between

eŒect of the experimental conditions on easiness of the

case, F(2,75) = 6.74, P < 0.01. As shown in ® gure 4,

cases were judged to be easier in the expert system with

production rules condition. Especially subjects with a

high need for cognition thought cases to be easier in the

production rule condition.

3.4. Computer experience

In a 2 (computer experience) ´ 3 (condition) repeated

measurement M ANOVA, computer experience did not

have an eŒect on compliance, objectivity/rationality, case

easiness, or authority. Thus, computer experience did not

aŒect user attitudes towards expert system advice.

4. Conclusions and discussion

Given the same advice, subjects thought an expert

system to be more objective and rational than a human

adviser, especially when the expert system advice was

J. J. Dijkstra et al.160

Figure 2. The eŒect of need for cognition on compliance

(top) and on authority (bottom).

Figure 3. Collapsed scores on objectivity/rationality (top)and the eŒect of need for cognition on objectivity/rationality

(bottom).

Figure 4. Collapsed scores on easiness of the case.

Dow

nloa

ded

by [

Yor

k U

nive

rsity

Lib

rari

es]

at 0

2:00

22

Nov

embe

r 20

14

given in a production rule format. Thus, H1 and H2 hold

for the objectivity and rationality construct. Objectivity

and rationality seem to be persuasive cues (beliefs) that

can make users accept advice of expert systems without

examining it. Experimental conditions did not seriously

aŒect scores on other, interpersonal, persuasiveness

factors (compliance and authority); only on compliance

there was a small within eŒect of condition by case.

Thus, H1 and H2 were not supported for compliance and

authority, which indicates that these factors are

relatively unimportant persuasive expert system cues.

The experimental conditions did have an eŒect on

perceived easiness of the cases. Surprisingly enough,

subjects found cases to be easier when the advice of the

expert system was given in a production rule style. Thus,

a case was thought to be less di� cult when the advice

was more di� cult to understand.

The experiment shows that studying an expert system

merely as a persuasive message source can give an

interesting new view on expert system use. The fact that

people are told that information comes from an expert

system is enough to in¯ uence their evaluation of the

information. This might explain why users of expert

systems sometimes neglect the incompetence of the

expert system. It is very likely that actual interaction

with an expert system will strengthen this persuasive

eŒect. However, more research has to be done to

examine this hypothesis.

The eŒect of need for cognition was diŒerent than

expected: H3 was not supported. It was hypothesized

that subjects with a low need for cognition would favour

the expert system advice more than subjects with a high

need for cognition. However, the eŒects of need for

cognition on the authority and compliance scores

indicate that people with a low need for cognition do

not like expert systems. A signi® cant correlation

between computer experience and need for cognition

supports this interpretation. This does not mean that

these people will disagree with the expert system.

According to Todd and Benbasat (1992, 1994a, b),

reduction of cognitive eŒort is a reason to accept expert

system advice. Most likely, people with a low need for

cognition want to reduce cognitive eŒort and, therefore,

accept advice of an expert system, even when they do

not like it.

The cases were thought to be easier in the production

rule condition, especially by subjects with a high need

for cognition. It seems likely that these subjects put an

eŒort in trying to understand the production rules and,

unfortunately, neglect the di� culties of the cases

involved. This implies that expert system use can

decrease attention paid towards problems. This would

make need for cognition an interesting but di� cult

factor when studying the attitudes towards expert

systems. People with a low need for cognition may be

uncritical towards expert system advice because they

want to reduce cognitive eŒort. People with a high need

for cognition may be uncritical because they pay too

much attention to the argument of the expert system and

forget to think about the case. Personal concern and

responsibility, two other factors that in¯ uence a user’ s

motivation to examine the advice (Petty and Cacioppo

1986a, b), will probably aŒect judgements on expert

system advice in a similar way. This makes these factors

interesting aspects for further research on attitudes of

expert system users.

References

ALDAG , R. J. and POWER, D. J. 1986, An empirical assessmentof computer-assisted decision analysis. Decision Sciences, 17,

572 ± 588.BARKI, H. and HARTW ICK, J. 1989, Rethinking the concept of

user involvement. Management Information Systems Quar-

terly, 13, 53 ± 63.

BARKI, H. and HARTW ICK, J. 1994, Measuring user participa-tion, user involvement, and user attitude. Management

Information Systems Quarterly, 18, 59 ± 82.BAROUDI, J. J., OLSON , M. H. and IVES, B. 1986. An empirical

study of the impact of user involvement on system usage andinformation satisfaction. Communications of the ACM , 29,

232 ± 238.BERG , M. 1995, Rationalizing medical work, Ph.D. disserta-

tion, University of Limburg.BERRY, D. C. and BROA DBEN T, D. E. 1987, Expert systems and

the man-machine interface, part two. Expert Systems, 4,

18 ± 28.

BODEN, M. 1990, Arti® cial intelligence and images of man, inR. Ennals and J. Gardin (eds), Interpretation in the

Humanities: Perspectives from Arti® cial Intelligence (Cam-bridge: Cambridge University Press), 10 ± 21.

BRIDGES, S. M. 1995, A model for justi® cation production byexpert planning systems. International Journal of Human-

Computer Studies, 43, 593 ± 619.CACIOPPO, J. T. and PETTY , R. E. 1982, The need for cognition.

Journal of Personality and Social Psychology, 42, 116 ± 131.CACIOPPO, J. T., PETTY , R. E. and KAO, C. F. 1984, The e� cient

assessment of need for cognition. Journal of Personality

Assessment, 48, 306 ± 307.

CHEN , Z. 1992, User responsibility and exception handling indecision support systems. Decision Support Systems, 8, 537 ±

540.COR NFORD, T., DOUKIDIS, G. I. and FORSTER, D. 1994,

Experience with a structure, process and outcome frame-work for evaluating an information system. Omega, 22,

491 ± 504.

DAVIS, F. D. 1989, Perceived usefulness, perceived ease of use,and user acceptance of information technology. Manage-

ment Information Systems Quarterly, 13, 319 ± 340.

DAVIS, F. D. 1993, User acceptance of information technology:System characteristics, user perceptions and behavioral

impacts. International Journal of Man-Machine Studies, 38,

475 ± 487.DAVIS, F. D., BAGOZZI, R. P. and WARSHAAW, P. R. 1989, User

acceptance of computer technology: A comparison of two

theoretical models. Management Science , 35, 982 ± 1003.

Persuasiveness of expert systems 161D

ownl

oade

d by

[Y

ork

Uni

vers

ity L

ibra

ries

] at

02:

00 2

2 N

ovem

ber

2014

DAVIS, F. D., LOHSE, G. L. and KOTTEM ANN, J. E. 1994,Harmful eŒects of seemingly helpful information on

forecasts of stock earnings. Journal of Economical Psychol-

ogy, 15, 253 ± 267.

DIJKSTRA, J. J. 1995, The in¯ uence of an expert system on theuser’ s view: How to fool a lawyer. New Review of Applied

Expert Systems, 1, 123 ± 138.GALLETTA , D. F. and LEDERER , A. L. 1989, Some cautions on

the measurement of user information satisfaction. Decision

Sciences, 20, 413 ± 438.

GATAIN, A. W. 1994, Is user satisfaction a valid measure ofsystem eŒectiveness? Information & Management, 26, 119 ±

131.GINZBERG , M. J. 1981, Early diagnosis of MIS implementation

failure: Promising results and unanswered questions. Man-

agement Science, 27, 459 ± 478.

HAR TW ICK, J. and BARKI, H. 1994, Explaining the role of userparticipation in information system use. Management

Science , 40, 440 ± 465.HOLLNAG EL, E. 1989, The reliability of expert systemsÐ an

inquiry of the background, in E. Hollnagel (ed.), The

Reliability of Expert Systems (Sussex: Ellis Horwood), 14 ±32.

IGBARIA, M. 1993, User acceptance of microcomputer technol-

ogy: An empirical test. Omega, 21, 73 ± 90.

IGBARIA, M. and NACH MAN , S. A. 1990, Correlates of usersatisfaction with end user computing: An exploratory study.Information & Management, 19, 73 ± 82.

IGBARIA, M., SCHIFFM AN, S. J. and W IECKOW SKI, T. J. 1994, The

respective roles of perceived usefulness and perceived fun inthe acceptance of microcomputer technology. Behaviour &

Information Technology, 13, 349 ± 361.IIVARI, J. and ERVA STI, I. 1994, User information satisfaction:

IS implementability and eŒectiveness. Information & Man-

agement, 27, 205 ± 220.

IVES, B., OLSON , M. H. and BAROUDI, J. J. 1983, Themeasurement of user information satisfaction. Communica-

tions of the ACM , 26, 785 ± 793.KEIL, M., BER ANEK, P. M. and KONSYNSKI, B. R. 1995,

Usefulness and ease of use: Field study evidence regardingtask considerations. Decision Support Systems, 13, 75 ± 91.

KIERS, H. A. L. 1990, SCA User’ s Manual (Groningen: iecProGAMMA).

KIERS, H. A. L. and TEN BERG E, J. M. F. 1989, Alternatingleast square algorithms for simultaneous components

analysis with equal weight matrices in two or morepopulations. Psychometrika, 54, 467 ± 473.

KOTTEM ANN , J. E. and REM US, W. E. 1987, Evidence andprinciples of functional and dysfunctional DSS. Omega, 15,

135 ± 143.KOTTEM ANN , J. E., DAVIS, F. D. and REM US, W. E. 1994,

Computer-assisted decision making: Performance, beliefsand the illusion of control. Organizational Behavior and

Human Decision Processes, 57, 26 ± 37.MELONE , N. P. 1990, A theoretical assessment of the user-

satisfaction construct in information systems research.Management Science, 36, 76 ± 91.

MOLENAA R, I. W., DEBETS , P., SIJTSM A, K. and HEMK ER, B. T.1994, MSP User’ s Manual (Groningen: iec ProGAMMA).

O’KEEFE, D. J. 1990, Persuasion: Theory and Research (New-bury Park: SAGE Publications).

PAREÂ , G. and ELA M, J. J. 1995, Discretionary use of personalcomputers by knowledge workers: Testing of a social

psychology theoretical model. Behaviour & Information

Technology, 14, 215 ± 228.

PETTY , R. E. and CACIOPPO, J. T. 1986a, Communication and

Persuasion, Central and Peripheral Routes to Attitude

Change (New York: Springer Verlag).PETTY , R. E. and CACIOPPO, J. T. 1986b, The elaboration

likelihood model of persuasion, in L. Berkowitz (ed.),Advances in Experimental Social Psychology, Vol. 19 (New

York: Academic Press), 123 ± 205.PIETER S, R. G. M., VERPLANKEN, B. and MODDE, J. M. 1987,

`Neiging tot nadenken’ : Samenhang met beredeneerdgedrag. Nederlands Tijdschrift voor de Psychologie, 42, 62 ±

70.PRIESTER, J. R. and PETTY , R. E. 1995, Source attributions and

persuasion: Perceived honesty as a determinant of message

scrutiny. Personality and Social Psychology Bulletin, 21,

637 ± 654.ROBEY , D. 1979, User attitude and management information

system use. Academy of Management Journal, 22, 527 ± 538.

SRINIVASAN, A. 1985, Alternative measures of system eŒective-ness: Associations and implications. Management Informa-

tion Systems Quarterly, 9, 243 ± 253.

SUH, C. and SUH, E. 1993, Using human factor guidelines for

developing expert systems. Expert Systems, 10, 151 ± 156.SZAJNA, B. 1993, Determining information system usage: Some

issues and examples. Information & Management, 25, 147 ±154.

TODD, P. and BENBASAT, I. 1992, The use of information in

decision making: An experimental investigation of the

impact of computer-based decision aids. Management

Information Systems Quarterly, 16, 373 ± 393.

TODD, P. and BENBASAT, I. 1994a, The in¯ uence of decision aidson choice strategies: An experimental analysis of the role of

cognitive eŒort. Organizational Behavior and Human Deci-

sion Processes, 60, 36 ± 74.

TODD, P. A. and BEN BA SAT, I. 1994b, The in¯ uence of decisionaids on choice strategies under conditions of high cognitive

load. IEEE Transactions on Systems, Man, and Cybernetics,24, 537 ± 547.

WALD ROP, M. M. 1987, Man-made Minds: The Promise of

Arti® cial Intelligence (New York: Walker and Company).

W ILL, R. P. 1992, Individual diŒerences in the performanceand use of an expert system. International Journal of Man-

Machine Studies, 37, 173 ± 190.WOGNUM , P. M. 1990, Explanation of Automated Reasoning:

How and Why?, Ph.D. dissertation, University of Twente.WORCHEL, S., COOPER, J. and GOETH ALS, G. R. 1988, Under-

standing Social Psychology (Chicago: Dorsey Press).YE, L. R. and JOHNSON, P. E. 1995, The impact of explanation

facilities on user acceptance of expert systems advice.Management Information Systems Quarterly, 19, 157 ± 172.

J. J. Dijkstra et al.162D

ownl

oade

d by

[Y

ork

Uni

vers

ity L

ibra

ries

] at

02:

00 2

2 N

ovem

ber

2014

Appendix

Law case wordings used in the expert system with

natural language argument (the original text was in

Dutch).

Tenancy case

About four years ago, Karel Biegman (age 27) moved in

with his mother again, after having lived on his own for 13

months. Mrs. Biegman, a widow, rents her house from the

rental company `Sun house,’ but Karel pays the rent. Mrs.

Biegman is disabled and rather old; soon she will move to

a Senior Citizens’ Home. Karel would like to be the new

tenant of his mother’ s house but the rental company does

not agree. Karel thinks he has a right to stay and looks for

advice. Therefore, at City Hall, Karel consults the expert

system `Tenant Advice.’ After Karel has answered some

questions on the case, the computer concludes:

As a non-tenant you want to claim tenancy rights, thus

you want to plead for joint tenancy (section 1623 Civil

Law).

To claim joint tenancy according to section 1623 Civil Law

you should have established a long-term household with the

current tenant. A parent and child cannot set up a long-

term household together, because, for the law, a child is

thought to live with its parents only temporarily (DeWit vs

Koning 1972). Usually, a child leaves home when he or she

is grown up. However, there is the possibility that, after

living on your own, you move in with your parents again. In

such a case, it is possible to set up a long-term household

together; that is when you really share the housekeeping

and you plan to live with your parents permanently

(Eleveld vs DeHaan 1981). Unfortunately, the period of

13 months you have lived on your own is too short to claim

joint tenancy after moving in with your mother again.

A bit depressed Karel goes home. He cannot believe that

the period he has lived on his own in¯ uences his tenancy

rights. Two months later the rental company orders

Karel to leave the house.

In the human adviser condition, the last two sentences

of the ® rst paragraph read:

`Therefore, at City Hall, Karel consults Mr.

Stienstra for advice. After Karel has answered

some questions on the case, Mr. Stienstra con-

cludes:

In the expert system with production rules condition the

case material reads:

Tenancy case

About four years ago, Karel Biegman (age 27) moved in

with his mother again, after having lived on his own for

13 months. Mrs. Biegman, a widow, rents her house

from the rental company `Sun house,’ but Karel pays

the rent. Mrs. Biegman is disabled and rather old; soon

she will move to a Senior Citizens’ Home. Karel would

like to be the new tenant of his mother’ s house but the

rental company does not agree. Karel thinks he has a

right to stay and looks for advice. Therefore, at City

Hall, Karel consults the expert system `Tenant Advice’ .

After Karel has answered some questions on the case,

the computer concludes:

A bit depressed Karel goes home. He cannot believe that

the period he has lived on his own in¯ uences his tenancy

rights. Two months later the rental company orders

Karel to leave the house.

Persuasiveness of expert systems 163

GOAL joined_tenancy {section 1623 Civil Law}{Comment: a non-tenant wants to claim tenancy rights}

RULE long-term-household_parent_childIF child_living_with_parent= TRUE

THEN long-term_household:= FALSE

{Source: DeWit vs Koning (1972). A parent and childcannot set up a long-term household together, because,

for the law, a child is thought to live with its parentsonly temporarily. Usually, a child leaves home when he

or she is grown up.}

INFERENCE child_living_with_parent= TRUETHUS long-term_household=

FALSE

RULE exception_long-term-household_parent_childIF

AND

AND

AND

THEN

child_has_lived_on_its_own= TRUE

period_child_living_in_its_own

> 2 {year}

parent_child_share_housekeeping=TRUE

intention_living_together_permanently= TRUE

long-term_household:= TRUE{Source: Eleveld vs DeHaan, 1981}

INFERENCE period_child_living_on_its_own< 2

THUS long-term_household=FALSE

RULE joined_tenancy

IFTHEN

ELSE

long-term_household= TRUEjoined_tenancy:= TRUE

joined_tenancy:= FALSE

INFERENCE long-term_household= FALSE

THUS joined_tenancy:= TRUE

CONCLUSION: you cannot claim joint tenancy

Dow

nloa

ded

by [

Yor

k U

nive

rsity

Lib

rari

es]

at 0

2:00

22

Nov

embe

r 20

14