A Multivariate Analysis of FactorsInfluencing Farm Machinery

Purchase Decisions

Thomas G. Johnson, William J. Brown andKevin O'Grady

This paper presents a model of the farm management process. The model suggests thatcertain socioeconomic characteristics of farm managers will influence their decision-makingprocess. Several characteristics are hypothesized and tested using multivariate techniques(multivariate analysis of variance, range tests, and multiple comparisons). The analysis indicatesthat the soil zone, value of machinery inventory, operator's age, and operator's educationinfluence the importance placed on each of 20 factors. On the basis of the analysis it wasconcluded that such a model of the farm management process can contribute to an understand-ing of farm management decisions. In addition, it was concluded that farm managers, farmmachinery dealers, and extension agents had significantly different perceptions of the impor-tance of these factors to farm managers. This latter conclusion suggests that more researchrelated to the actual process of decision making is warranted.

The selection of machinery that is suit-able and profitable for their particularfarm business is a recurrent, complex, andimportant decision confronting farm busi-ness managers. A conceptual model of themanagement process is presented that in-cludes a criteria-based decision analysisintroduced as a complement to neoclassi-cal microeconomic theory. In a survey offarm business managers, machinery deal-ers, and agricultural representatives (ex-tension agents), farmer respondents wereasked to rate the importance of variousfactors in making machinery purchase de-cisions, and machinery dealer and agri-cultural extension agent respondents wereasked to rate the importance of variousfactors to farmers making machinery pur-chase decisions. Multivariate analysis pro-

Thomas G. Johnson is Assistant Professor in the De-partment of Agricultural Economics at VirginiaPolytechnic Institute and State University. WilliamJ. Brown and Kevin O'Grady are Assistant Professorand Graduate Research Assistant at the University ofSaskatchewan, respectively.

The authors with to thank anonymous reviewers fortheir useful comments on an earlier draft.

cedures are used to test whether certaincharacteristics of the farm and the farmbusiness manager have an effect on theimportance attributed to factors affectingfarm machinery purchase decisions, andwhether machinery dealers and agricul-tural extension agents differ significantlyfrom farm business mangers in their rat-ing of the importance of these same fac-tors.

The objectives of the research were to:1) test the relative importance of varioussocioeconomic characteristics on the de-cision to purchase machinery, and 2) todetermine how accurately farm machin-ery dealers and agricultural extensionagents understand the decision-makingprocesses of farmers.

Neoclassical Theory andDecision-Making Models

Neoclassical microeconomic theoryproposes to predict the behavior of deci-sion makers under a variety of circum-stances, yet, by itself, it is lacking as abasis for predicting or even understanding

Western Journal of Agricultural Economics, 10(2): 294-306© 1985 by the Western Agricultural Economics Association

Machinery Purchase Decisions

routine farm management decisions. Tounderstand the purchase of farm machin-ery and other day-to-day decisions of farmmanagers, neoclassical microeconomictheory should be supplemented with avastly different approach. This approachmust consider a number of questions."How large a machine should be pur-chased?" "Should it be new or used?""What special features should it have?""When should it be replaced?" And, per-haps most importantly: "What will myfriends and neighbors think of my deci-sion?"

Effective management of a commercialfarm requires the majority of these deci-sions to be correct. Farm managers re-quire access to both information and aprocess. The information (or content ofmanagement) includes the myriad oftechnical, biological, economic, and socio-logical data related to a modern agricul-tural enterprise. The management processis the implicit or explicit method used bythe manager to assimilate the informationand arrive at an end, usually the accom-plishment of predetermined goals. Allmanagers, whether aware of it or not, nec-essarily employ a management process.The sophistication and nature of this pro-cess varies widely from one manager toanother. Poor management results can oc-cur because of inadequate content, an in-adequate process, or some combination ofboth.

A management process outlined byCromier, Mitchel, and McGiffin (based onconcepts developed by Kepner and Tre-goe) is a model of an actual managementprocedure which conveniently traversesthe gap between microeconomic theoryand real world situations. Figure 1 depictsthe major components of this model, whichinclude issue analysis, problem (opportu-nity) analysis, decision analysis, and actionplanning or potential problem analysis. Is-sue analysis is the usual starting place forthe process and encompasses the formu-lation of business and personal goals and

ISSUEI

PROBLEMI

ALTERNATIVE

DECISION I

Figure 1.cess.

.* . .. *ISSUEi

PROBLEMj

ALTERNATIVEk

DECISION m

ACTIONS

A Model of the Management Pro-

the prioritization of issues. Problem (op-portunity) analysis is the process of rec-ognizing problems, specifying the what,where, when, and extent of the problem,analyzing for distinctions and changes, andfinding and testing the cause. Decisionanalysis is the process of setting and clas-sifying criteria, comparing and choosingamong available alternatives based on thecriteria, and assessing the adverse conse-quences associated with the choice. Final-ly, potential problem analysis and actionplanning outline the procedures to be tak-en to insure that decisions and problemsare acted upon and that goals are met.This procedure includes the anticipationof potential problems and their possiblecauses, the taking of preventative action,and, in case this fails, the making of con-tingency plans. Each component of themanagement process can interact with anyone of the other three components at anygiven time. For example, new priority is-sues may arise when business and personalgoals change as a result of the problemsolution, potential problem analysis, or de-cisions being made. In addition, decisionsmay trigger new problems and actionplans and vice versa.

295

Johnson et al.

Western Journal of Agricultural Economics

The decision analysis component is ofparticular concern in this paper; there-fore, its description is expanded here. Themanager begins decision analysis by stat-ing as concisely as possible the decisionthat is to be made. For example, "Whattractor size and type is best for my farm-ing situation?" Next, the manager makesa list of the criteria upon which the de-cision will be based. Criteria usually in-clude power and time requirements, mod-el or make reputation, cash and creditconstraints, and technological and otheroptions. Some of these criteria are man-datory; others are desirable. Mandatorycriteria must be objective, realistic, andmeasurable whereas desirable criteria cansimply be statements of the manager'spreferred results. Desirable criteria arepersonal and do not have to be objective,realistic or measurable, but they do havea potential influence on the decisionchoice. Each alternative-size and makeof tractor, various financing arrange-ments-that meets the mandatory criteriais considered in terms of the desirable cri-teria. The desirable criteria are weightedaccording to importance, and each alter-native is scored for each desirable crite-rion. Which criteria are included, and theweighting of each, is itself a decision vari-able. Managers, as circumstances changeand as they acquire experience, will addand delete criteria and change the relativeweighting. The alternative with the high-est weighted score across all desirable cri-teria is the manager's preliminary choice.

Before making the final decision, themanager analyzes the risk involved. Heconsiders the most dangerous scenariopossible, and compares his preliminarychoice with perhaps two of the next bestalternatives. The dangers involved inchoosing any one of these top alternativesare assessed in terms of the perceivedprobability and severity of their occur-rence. For example, the first-chosen alter-native could be manufactured by a com-pany that has a high probability of going

TABLE 1. Factors a Farmer May ConsiderWhen Purchasing a Tractor orCombine.

1. Change in Size of the Farming Operation2. Time Available Due to Weather3. Time Available Due to Labor Supply4. Time Available Due to Desire for Leisure5. Soil Texture6. Topography7. Size of Other Machinery Already Being Used8. Old Machine Wearing Out9. New Model has Improvements not on Old Model

10. More Income Tax Deductions11. Money Available to Pay Cash12. Credit Available13. Custom Hiring of Machine Work for Others14. Fuel Efficiency15. Past Experiment Indicates the Benefits Outweigh

the Costs16. Mental Calculation Indicates the Benefits Out-

weigh the Costs17. Written Calculation Indicates the Benefits Out-

weigh the Costs18. Farm Records Indicate the Benefits Outweigh the

Costs19. Family Persuasion20. Friends' and Neighbors' Persuasion

bankrupt and the manager might be afraidof a future lack of spare parts. The finalchoice is based on the results of the pre-liminary choice and the risk analysis.

Hypotheses

The value of the above model is that itstresses the preparatory stages of manage-ment, and suggests that management de-cisions are based on several factors whoserelative importance varies among man-agers. Thus, it seems useful to gain a bet-ter understanding of the determinants ofthe factors used by farm managers in ma-chinery purchase decisions. In this study,farmers were asked to rate a series of fac-tors according to importance in their farmmachinery purchase decisions (Table 1),and machinery dealers and agriculturalrepresentatives were asked to rate the im-portance of the same factors to farmers.These factors are interpreted as a list ofpossible desirable criteria for farm ma-chinery purchases. The objective is to see

296

December 1985

Machinery Purchase Decisions

if these selected factors are actually im-portant in farm machinery purchase de-cision making and under what conditionsthat importance changes. The list does notinclude all possible criteria and does notcover the entire machinery purchase de-cision. Other decisions related to machin-ery purchases are not addressed: "Do Ineed a new machine in the first place?""What specific set of characteristics shouldthe maching have?" "From whom shouldI purchase the machine?" Also, each fac-tor on the list would most likely be re-worded in an actual decision analysis tomore accurately reflect the manager's de-sired results. For example, "change in sizeof farming operation" could be rewordedto "If I rented an additional 200 acres ofland, would the combine be large enoughto complete harvesting in good time?"

Each manager has a unique situationand conceptually weighs the various de-sirable factors differently when making afarm management decision. For example,some managers may feel that timeliness isof particular importance while, to others,cash flow considerations may be a greaterconcern. If farm managers maximize util-ity rather than profit, they may considercertain noneconomic factors important.

To the extent that the general charac-teristics of farms influence the economicforces acting on farm managers, thesesame forces may influence what farmersperceive as important considerations. Fol-lowing this logic, it is hypothesized thatthe importance that farmers place on var-ious considerations when purchasing farmmachinery is influenced by:

1) the soil zone in which the farm islocated;

2) the type of products produced on thefarm;

3) the size of the farm;4) the current value of the machinery

inventory;5) the operator's age; and6) the operator's education.

Furthermore, it is hypothesized thatmachinery dealers and agricultural rep-resentatives understand the decision-mak-ing process which farmers employ and canaccurately predict the factors that farmersconsider important.

The Data

The data for this study were obtainedfrom a mail survey undertaken in 1980[Brown and Strayer]. The survey samplewas drawn from those Saskatchewanfarmers who registered a farm truck witha Gross Vehicle Weight of 11,000 poundsor greater, because, in the authors' expe-rience, bona fide farmers almost alwaysown their trucks and it was desirable toeliminate hobby farmers from the sample.

The survey questionnaire listed a num-ber of factors which were hypothesized tobe important in a farmer's decision to pur-chase a tractor, combine, or both (see Ta-ble 1). The farmer was asked to rate theimportance of each factor in his decision-making process by responding with 1, 2,3, or 4. (A "1" signified not important anda "4," very important.) In addition to rat-ing these factors, the farmers were askedseveral questions with respect to the phys-ical and socioeconomic characteristics oftheir farm operation.

A total of 4,939 farmers was sent thequestionnaire in February 1980 and 1,482responded. Of these, 577 responses wererejected because they were incomplete,leaving 905 responses for use in this study.

To supplement the responses byfarmers, questionnaires were mailed to the405 Saskatchewan members of the Sas-katchewan-Manitoba Farm ImplementDealers Association and to all 42 Agricul-tural Representatives (extension agents) inthe Saskatchewan Department of Agri-culture. These extension agents and deal-ers were asked to rate the decision-makingcriteria in such a way as to describe whatthey perceived as most important/leastimportant to the farm operators. A total

297

Johnson et al.

Western Journal of Agricultural Economics

of 132 machine dealers and 30 extensionagents provided usable responses.

Multivariate Analysis

The purpose of this study was to testthe null hypothesis that certain character-istics of the farm and socioeconomic char-acteristics of the farm operator have noeffect on the importance that respondentsattribute to the decision-making factors.Simply stated, do farmers facing differentcircumstances place different weights onfarm machinery decision-making factors?While it would be rather straightforwardto test for significant differences amonggroups on the basis of a single variable,our conceptual model predicts that manyfactors are weighted simultaneously; thus,multivariate analysis is more appropriatethan the traditional univariate analysissince it considers the interdependencyamong these factors. A single multivariateanalysis with many dependent variablesincurs much less risk of committing a TypeI Error than do several univariate analyseswith one dependent variable each. Forboth heuristic and rigorous discussions ofthe appropriate applications of multivari-ate analysis, see Harris and Morrison.

In the first part of the analysis, six farmand socioeconomic characteristics aretreated as independent variables. These are(1) soil zone, (2) farm type, (3) farm size,(4) present value of farm machinery, (5)operator age, and (6) operator education.Each of these variables is discrete.

The first step is to determine if anyoverall relationship exists between the de-cision factors and the six independentvariables. Since all of the independentvariables are discrete, multivariate anal-ysis of variance (MANOVA) is most ap-propriate. For k = 6 discrete independenttreatments, a six-way MANOVA is per-formed. Such a test indicates the amountof variation in the dependent variables,explained by the k treatments. If one ofthe k treatments is age, for example,

MANOVA will indicate (at a given levelof significance) if an operator's age influ-ences his (her) system of weights.

Should any independent variable havea significant effect, then further analysisis required to determine which level orlevels of that independent variable aresignificantly different from each other. Tothis end simultaneous multivariate multi-ple comparisons (SMMC) are employed.This method, for example, will determine,should the farmer's age be proven signif-icant by the MANOVA, whether thosefarmers in any particular age group aresignificantly different from others in theiroverall rating of the decision-making fac-tors.

The strategy of SMMC in this analysisinvolves the performance of several one-way MANOVAs for a given significant in-dependent variable (such as age). TheMANOVAs are achieved through multi-variate regression with dummy variables.By successively removing different dum-my variables, F statistics can be calculatedfor the marginal contribution of each levelof the variable. If the F statistic is greaterthan the appropriate critical value, thenthe associated group or discrete value hasa significant effect on the decision-makingprocess. If the F statistic for farmers un-der 25 is significant, for example, the an-alyst can conclude that younger farmersmake decisions on the bases of differentfactors.

At this point the analysis will indicatewhich independent treatment variableshave a significant effect on the overallweighting, and, for those which are sig-nificant, which levels of the treatmenthave a significant effect on the overallweighting of factors. This knowledge initself tells a great deal about the factorsinfluencing an individual's decisions, butthe analyst may want to know if this sig-nificant effect on the overall weighting isfocused on any particular decision factoror group of factors. To this end, rangetests-univariate or multivariate-may be

298

December 1985

Machinery Purchase Decisions

TABLE 2 Number of Observations in EachLevel of the Six Independent Vari-ables.

Num-ber of

Re- Obser-Variable sponse Class vations

Soil Zone 1 Brown 2372 Dark Brown 2793 Black 389

Farm Type 1 Grain 5372 Mixed 3503 Livestock 18

Farm Size 1 <640 Acres 2162 640-1,279 Acres 3743 1,280-1,920 Acres 1924 >1,920 Acres 123

Value of 1 <$50,000 219Machinery 2 $50,000-$149,000 487Investment 3 $150,000-$249,000 158

4 $250,000-$350,000 295 > $350,000 12

Operator 1 <25 years 34Age 2 25-40 years 352

3 41-65 years 4934 >65 years 26

Operator 1 <6 years 15Education 2 7-9 years 240

3 10-12 years 4564 Technical Diploma 1085 University Degree 86

used on each of the decision-making fac-tors. The more tests run, the greater thechance that at least one leads to an incor-rect conclusion. Thus, for more than onetest, the univariate method underesti-mates the probability of a Type I Error inat least one test. The multivariate test con-siders the number of individual tests andis therefore less discerning.

Range tests construct a confidence in-terval (for a given confidence level) aroundthe mean value of the dependent variableassociated with each level of an indepen-dent variable. Should all mean values ofthe dependent variable fall within thesame confidence interval, then it followsthat the associated independent variabledoes not make its contribution through thisparticular dependent variable. For ex-

ample, if the comparisons above indicatethat age is a factor, and that youngerfarmers tend to make different decisions,then the range test can pinpoint whichparticular factors they consider more orless important. If the mean responses forall age groups fall into the same confi-dence interval, then it is possible to con-clude that age does not have its effect inthis factor (or factors). The multivariatetest employs a procedure detailed in Mor-rison [pp. 194-204] and Heck [p. 627].

As noted earlier, the survey also includ-ed a sample of machinery dealers and ag-ricultural extension agents. These twogroups were asked to rate the importanceof each factor in the farmer's decision-making process. Given these data, it ispossible to test the null hypothesis thatthere are no significant differences be-tween farmers, machinery dealers, andagricultural extension agents in their per-ception of the importance of each deci-sion-making factor in the farmer's deci-sion-making process. To test thishypothesis, a one-way MANOVA is per-formed in which the single treatment in-cludes three occupations-farmer, exten-sion agent, and machinery dealer.

Should the analysis indicate that any twoof the three occupational groups are sig-nificantly different in their factor ratings,then it will be of interest to determinewhich of the dependent variables contrib-ute to this difference. Here univariate andmultivariate range tests, as described pre-viously, are employed.

Analysis

Table 2 indicates the distribution of ob-servations among the various levels of thesix independent variables. Table 3 givesthe mean response to each factor for eachclass of each independent variable. Theobject of the following analysis is to de-termine if there are any statistical differ-ences among these means.

First, which, if any, of the six indepen-

299

Johnson et al.

Western Journal of Agricultural Economics

TABLE 3. Mean Responses to Twenty Factors by Levels of Independent Variables.

Value of~~~~~~~~ D e c l ~~~~~~~i- ~Machinery

sion Over- Soil Zone Farm Type Farm Size InvestmentFac- alltor Mean I II III I II III I II III IV I II

1 3.256 3.262 3.226 3.275 3.242 3.277 3.278 3.176 3.222 3.391 3.293 3.274 3.2402 3.204 3.093 3.211 3.267 3.233 3.169 3.056 3.056 3.195 3.292 3.358 3.059 3.1773 2.873 2.722 2.910 2.938 2.879 2.860 2.944 2.690 2.781 3.094 3.130 2.671 2.8584 1.830 1.865 1.896 1.763 1.858 1.791 1.778 1.759 1.869 1.771 1.935 1.721 1.8975 2.335 2.228 2.348 2.391 2.341 2.331 2.222 2.398 2.350 2.281 2.260 2.338 2.3496 2.512 2.540 2.513 2.499 2.467 2.591 2.389 2.458 2.529 2.589 2.447 2.502 2.5287 3.036 3.097 3.043 2.995 3.006 3.089 2.944 2.954 3.053 3.089 3.049 2.936 3.0578 3.316 3.401 3.287 3.285 3.291 3.340 3.611 3.255 3.307 3.344 3.407 3.324 3.3169 2.684 2.629 2.746 2.674 2.717 2.649 2.389 2.519 2.706 2.760 2.788 2.498 2.717

10 2.695 2.713 2.767 2.632 2.689 2.714 2.500 2.588 2.701 2.797 2.707 2.507 2.71911 2.879 2.873 2.860 2.987 2.847 2.920 3.056 3.060 2.920 2.729 2.675 3.023 2.88112 2.726 2.717 2.738 2.722 2.659 2.834 2.611 2.731 2.703 2.714 2.805 2.749 2.73513 1.715 1.722 1.760 1.679 1.670 1.783 1.722 1.750 1.717 1.714 1.650 1.749 1.72314 3.193 3.131 3.211 3.219 3.123 3.291 3.389 3.296 3.184 3.219 3.000 3.342 3.16815 3.121 3.097 3.204 3.077 3.110 3.129 3.333 3.056 3.072 3.240 3.203 3.073 3.06716 2.653 2.540 2.677 2.704 2.620 2.711 2.500 2.644 2.620 2.677 2.732 2.635 2.64917 2.897 2.945 2.889 2.874 2.898 2.889 3.056 2.912 2.848 2.906 3.008 2.918 2.86918 3.072 3.122 3.093 3.026 3.039 3.131 2.889 3.046 3.056 3.109 3.106 3.068 3.05719 1.843 1.835 1.817 1.866 1.797 1.897 2.167 1.931 1.770 1.922 1.789 1.836 1.86020 1.462 1.460 1.498 1.440 1.473 1.443 1.556 1.542 1.471 1.438 1.341 1.571 1.468

dent variables lead to a significant differ-ence in responses? Table 4 reports the re-sults of a six-way MANOVA that measuresthe differences in response due to farmand operator characteristics. All but farmsize and farm type have a significant in-fluence on responses at the 95 percent levelor better. It is important to recall that thisresult does not imply that farm type andfarm size do not influence which ma-chines are purchased or the amount ofmachines purchased. Rather, it suggeststhat the factors which decision makers takeinto consideration are the same for smalland large farms and for grain, mixed, andlivestock farms. This finding vividly dis-tinguishes the neoclassical decision model(which would consider only farm type,farm size, and machinery inventory), froma farm management process model.

Soil zone, value of machinery invento-ry, the operator's age and education alltend to influence the factors considered by

farmers. The next issue is which soil zones,which age groups, etc., lead to differentresponses. The results of the various one-way MANOVAs and the multiple com-parisons are reported in Tables 5 and 6.Table 5 indicates the F statistic related toeach pair of levels within an independentvariable. The numbers in brackets are theprobability that the two groups come fromthe same population. Table 6 summarizesthis information for three levels of signif-icance.

These tables indicate that the responsesof farmers in the brown soil zone differfrom those in the black zone, but that thoseof the dark brown zone are intermediateand thus indiscernible from either. At the95 percent significance level those opera-tors with less than $50,000 of machinery(category 1) and those with more than$350,000 of machinery (category 5) gavedistinct responses. Categories 2 and 3 canbe distinguished but category 4 is inter-

300

December 1985

Machinery Purchase Decisions

TABLE 3. Extended.

Value of MachineryInvestment Operator Age Operator Education

III IV V I II III IV I II III IV V

3.304 3.207 3.083 3.353 3.384 3.166 3.115 3.533 3.154 3.276 3.333 3.2913.443 3.276 3.667 2.941 3.241 3.205 3.038 3.400 3.242 3.213 3.102 3.1513.139 3.172 2.917 2.676 2.949 2.832 2.885 2.933 2.829 2.864 2.861 3.0471.810 1.793 1.500 2.088 1.855 1.813 1.500 1.400 1.788 1.833 1.861 1.9772.253 2.414 2.583 2.088 2.207 2.438 2.423 2.400 2.617 2.296 2.185 1.9032.462 2.690 2.417 2.529 2.403 2.604 2.269 2.200 2.663 2.509 2.556 2.1283.095 3.241 2.750 2.971 3.037 3.037 3.115 3.400 3.038 3.061 2.944 2.9533.266 3.414 3.583 3.618 3.330 3.292 3.192 3.000 3.329 3.303 3.370 3.3372.810 2.931 2.500 2.794 2.668 2.690 2.654 2.533 2.817 2.662 2.611 2.5472.791 3.069 3.000 2.941 2.580 2.777 2.385 2.800 2.821 2.700 2.639 2.3722.665 2.966 2.833 2.824 2.798 2.939 2.923 3.267 3.033 2.866 2.769 2.5932.595 3.276 2.333 3.147 2.733 2.704 2.500 2.333 2.850 2.754 2.574 2.4881.614 1.759 2.000 2.000 1.682 1.728 1.538 1.467 1.804 1.732 1.630 1.5233.051 3.414 2.833 3.118 3.148 3.219 3.423 3.667 3.417 3.163 3.019 3.0123.310 3.379 3.167 3.000 3.037 3.183 3.269 3.267 3.229 3.116 3.102 3.8492.646 3.034 2.333 2.706 2.696 2.617 2.692 2.867 2.796 2.629 2.565 2.4532.842 3.345 3.333 3.088 2.980 3.826 2.846 2.667 2.900 2.895 2.917 2.9193.057 3.276 3.417 3.294 3.000 3.101 3.192 3.200 3.121 3.105 2.954 2.8841.753 2.138 1.750 1.853 1.741 1.913 1.885 1.667 2.029 1.811 1.741 1.6511.335 1.310 1.333 1.647 1.477 1.436 1.538 1.533 1.533 1.439 1.389 1.477

mediate and not distinguishable fromeither. With respect to age, farmers under25 years (category 1) are significantly dif-ferent from all others at the 95 percentconfidence level. Those 25 to 40 are dis-tinguishable from those 41 to 65, butfarmers over 65 are not distinguishablefrom either the 25 to 40 or the 41 to 65age group. Finally, with respect to edu-cation level, each group is significantly dif-ferent from all others except those with atechnical diploma (category 4), which isan intermediate between those below andabove.

Thus there is a considerable amount ofregularity in the responses due to relation-ships between the responses and charac-teristics of the farm and farm operator.While it is difficult to impute any ordinalrelationships when a multivariate tech-nique is employed, the analysis seems toindicate a linear gradation of responses assoil zone and education change, and a cur-

vilinear (U shaped) change as age and val-ue of machinery change.

Next, multivariate confidence intervalsare calculated for individual decision-making factors to determine which fac-tors, if any, play a particularly importantrole in the observed multivariate relation-ships. Table 7 compares 95 percent con-fidence intervals with the mean responses

TABLE 4. F-Values and Prob-Values from Six-Way MANOVA.

Independent Probability ofVariable Treatment F-Value Type I Error

Soil Zone 1.42 0.0449*Farm Type 1.27 0.1211Farm Size 1.22 0.1212Value of Machinery

Inventory 1.79 0.0001**Operator Age 1.81 0.0002**Operator Education 1.75 0.0001**

* Significant at the 95 percent confidence level.** Significant at the 99 percent confidence level.

301

Johnson et al.

Western Journal of Agricultural Economics

TABLE 5. F-Values from Pairwise Simultaneous Multivariate Multiple Comparisons, for Inde-pendent Variables (Soil Zone, Machinery Inventory, Age, and Education).

ClassIndependent ClVariable Class 2 3 4 5

Soil Zone 1 1.10 2.16**(.3397) (.0023)

2 1.06(.3923)

Value of Machinery 1 2.29** 5.33** 2.00** 1.80**Inventory (.0011) (.0001) (.0057) (.0170)

2 2.82** 1.38 1.63*(.0001) (.1248) (.0390)

3 1.45 1.44(.0899) (.0947)

4 1.63*(.0389)

Operator Age 1 1.64* 2.05** 2.14**(.0376) (.0043) (.0026)

2 4.01** 1.30(.0001) (.1710)

3 1.04(.4137)

Operator Education 1 1.67* 1.80* 2.18** 2.27**(.0336) (.0167) (.0020) (.0012)

2 2.47** 2.63** 4.68**(.0004) (.0001) (.0001)

3 0.83 2.40**(.6765) (.0006)

4 1.25(.2044)

* Significant at the 95 percent confidence level.** Significant at the 99 percent confidence level.

(The numbers in parentheses are the probability of a Type I Error, i.e., the Prob-Value.)

to selected factors.1 It is clear that none ofthese factors generate significantly differ-ent means on an individual basis. Thus itappears that the influence of the four in-dependent variables, while significant, isoperative in an overall, multivariate senseonly.

Table 7 indicates those factors which

In order to reduce the calculations, the total set offactors was reduced to this smaller set by selectingonly those factors with significantly different meanson the basis of the Duncan's Multiple Range Test-a univariate test. Since the multivariate test is lessdiscerning, those differences which are insignificant

'in the univariate test cannot be significant in themultivariate case.

are relatively (if not absolutely) signifi-cant. Consider, for example, those factorswith class mean differences (column 4)that are at least 40 percent of the multi-variate confidence interval (column 5). 2

Soil zone differences made their biggest

2 Since none of the criteria in Table 7 are significantin multivariate tests and all are significant in theunivariate test, the dangers inherent in the simplertest are indicated. On the other hand, the moredemanding multivariate test tends to overlook cer-tain weak but, perhaps, meaningful differences. Thelist in Table 7 indicates which criteria merit furtherresearch. The 40 percent factor was chosen arbi-trarily in order to identify the most significant fromthis group.

302

December 1985

Machinery Purchase Decisions

TABLE 6. Distinct Groupings on the Basis ofSimultaneous Multivariate MultipleComparisons.

Independent Confidence LevelIndependentVariable Class 99% 95% 90%

Soil Zone 1 A A A2 AB AB AB3 B B B

Value of 1 A A AMachinery 2 B C CInventory 3 C B D D

4 BC CD C5 BC B B

Operator 1 A A AAge 2 AB B B

3 C C C4 BC BC BC

Operator 1 AB A AEducation 2 A B B

3 BC C C4 CD CD CD5 D D D

Note: Levels, within an independent variable, with thesame letter cannot be distinguished at the associ-ated level of confidence. Those with two or moreletters cannot be distinguished from two or moreother levels. For example, at the 90 percent confi-dence level, Operator Age, class 1(A) is discerniblefrom all others, but class 2(B) is not discernible fromclass 4 and class 4(C) is not discernible from class3. Classes 2 and 3 are discernible, suggesting thatclass 4 falls between classes 2 and 3.

impact on responses related to (2) timeavailable due to weather, and (3) timeavailable due to labor supply. The currentvalue of machinery most affected the re-sponses to (10) income tax deductions, and(11) cash availability. Age appears to bemost influential in its effect on (19) familypersuasion. Finally, the operator's educa-tion has its largest impact on (5) soil tex-ture, (6) topography, and (10) income taxdeductions. It is important to note that thediscussion above is not intended to implythat farmers consider the above factorsmore important than others. Rather it in-dicates that the importance ratings variedmore than usual in response to changes inthe independent variables. For example,family persuasion is rated quite low by allgroups, but those farm managers between

TABLE 7. Multivariate Confidence IntervalsAssociated with Class Means forIndependent Variables.*

Multi-variateConfi-

Decision- denceMaking Differ- Interval

Criterion Highest Lowest ence (95%)

23

16

Value2349

101112161720

14568

1219

14569

101112131419

Soil Zone Class Mean3.267 3.093 0.1742.938 2.722 0.2162.704 2.540 0.164

0.4210.5210.477

of Machinery Inventory Class Mean3.6673.1722.5832.9313.0693.0233.2763.0343.3452.292

Operator3.3842.0882.4382.6043.6183.1471.647

3.059 0.6082.671 0.5012.253 0.3302.498 0.4332.507 0.5622.665 0.3582.333 0.9432.333 0.7012.842 0.5031.966 0.326

Age Class Mean3.115 0.2691.500 0.5882.088 0.3502.269 0.3353.192 0.4262.500 0.6471.436 0.211

Operator Education Class Mean3.5331.9772.6172.6632.8172.8213.2672.8501.8043.6671.633

3.1541.4001.9302.1282.5332.3722.5932.4881.4673.0121.333

0.3790.5770.6870.5350.2840.4490.6740.3620.3370.6550.300

1.6551.3701.7371.2211.3350.7142.4402.1931.2870.996

1.0661.4831.1771.3141.3441.7800.418

1.4661.6640.8610.8531.6510.8501.9101.8941.6351.7190.783

* The confidence intervals are reported only for thosedecision-making factors which showed significantclass differences at the 95 percent univariate confi-dence level.

the ages of 25 and 40 tended to rate itlower than all other age groups.

The final stage of the analysis involvesthe testing for correspondence between theresponses of farmers, agricultural repre-

303

Johnson et al.

Western Journal of Agricultural Economics

TABLE 8. Ranked Mean Importance Rating by Respondent Occupation*

Decision- Farmers Machinery Dealers Ext. AgentsMakingFactor Mean Rank Mean Rank Mean Rank Overall Mean

8 3.319 1 3.018 7 2.920 7 3.2781 3.261 2 3.442 1 3.320 1 3.2812 3.201 3 3.239 4 3.080 4 3.203

14 3.190 4 2.965 8 2.240 15 3.14415 3.119 5 3.044 6 2.720 9 3.10218 3.070 6 2.841 10 2.040 18 3.0227 3.038 7 2.770 12 3.000 6 3.009

17 2.892 8 2.823 11 1.960 19 2.8633 2.873 9 3.257 3 3.240 3 2.921

11 2.872 10 2.416 16 2.360 12 2.81212 2.725 11 2.920 9 3.000 5 2.75210 2.692 12 3.381 2 3.280 2 2.7789 2.684 13 3.044 5 2.800 8 2.724

16 2.651 14 2.673 14 2.640 10 2.65313 2.613 15 2.300 18 1.800 20 2.5516 2.516 16 2.327 17 2.320 13 2.4925 2.343 17 2.159 19 2.080 17 2.317

19 1.841 18 2.717 13 2.280 14 1.9434 1.830 19 2.027 20 2.200 16 1.859

20 1.461 20 2.540 15 2.560 11 1.599

* Overall mean may not equal the sum of the weighted means due to rounding error.

sentatives, and machine dealers. Table 8presents the mean responses of the threegroups to the 20 factors, ordered accord-ing to the farmer's ranking. Thus factor8, size of other machinery, was rankedhighest by farmers but seventh highest bymachinery dealers and agricultural rep-resentatives.

A one-way MANOVA performed onthese data indicates extremely significantdifferences among the three groups(greater than 99.8 percent confidence forall comparisons). Furthermore, calcula-tion of 95 percent multivariate confidenceintervals around individual factors (seeTable 9) indicates that three variables wereparticularly important in differentiatingthe groups. These were (10) more incometax deductions, (19) family persuasion, and(20) persuasion by friends and neighbors.In all three cases, farmers rated the factorslower than did machinery dealers and ag-ricultural extension agents.

Thus it would appear that farm ma-chinery dealers and agricultural represen-

tatives-two groups who are vitally con-cerned with decisions made by farmers-could learn much from a study such asthis.

Conclusions

The analysis above indicates the com-plexity of the farm management decisionprocess. While few, if any, individual de-cision factors stand out as being related tothe decision-makers circumstances (age,education, etc.), it is equally clear thatthese factors taken together are stronglyinfluenced by the circumstances underwhich they are made. Thus the value ofmultivariate analysis is also illustrated.

Based on this study we are unable toreject the null hypotheses that farm typeand farm size have no bearing on the im-portance attributed to the factors consid-ered when buying a tractor or a combine.It is not suggested that these variables haveno effect on the decisions themselves. Onthe contrary, microeconomic theory pre-

304

December 1985

Machinery Purchase Decisions

TABLE 9. Multivariate Confidence IntervalsAssociated with Class Means forRespondent Occupation.*

MultivariateDecision- lass Mean ConfidenceMaking Differ- IntervalCriterion Highest Lowest ence (95%)

1 3.442 3.320 0.122 0.4873 3.257 2.873 0.384 0.6074 2.200 1.830 0.370 1.0817 3.038 2.770 0.268 0.5098 3.319 2.920 0.399 0.9949 3.044 2.684 0.360 0.550

10 3.381 2.692 0.689 0.591**11 2.872 2.360 0.512 1.23314 3.190 2.240 0.950 1.13315 3.119 2.720 0.399 0.97817 2.892 1.960 0.932 1.15718 3.070 2.040 1.030 1.15619 2.717 1.841 0.876 0.556**20 2.560 1.461 1.099 0.934**

* This analysis involves 1,081 observations. The con-fidence intervals are reported for only those deci-sion-making factors which showed significant classdifference at the 95 percent univariate confidencelevel.

** Significant difference between class with the high-est mean and classes with the lowest mean.

diets that these variables will be very im-portant in machinery purchase decisions.As pointed out above, the fact that thesevariables are not significant elements inthe farm management process under-scores the difference (and the comple-mentarity) between the neoclassical andfarm management process models of be-havior. The two models address differentaspects in the overall management pro-cess. Production theory relates to the op-timal mix and size of the machinery in-ventory. The process model indicates howthe decisions are made.

The analysis allows us to reject the nullhypotheses that soil zone, value of ma-chinery inventory, the operator's age andeducation do not affect the importance ofthe various factors. Each of these, in fact,leads to patterns of responses which differaccording to the levels of the variables.Farmers in the brown and black zone,perhaps because of differential growing

seasons, place different emphases on thetiming-related factors. Farmers in theblack soil zone rate them higher and thosein the brown soil zone rate them lower.Farmers with more education place lessimportance than others on soil texture, to-pography, and income tax deductions.

The value of machinery inventory andoperator age do not generate suchstraightforward patterns of response. Theimportance of income tax deductions in-creases as the value of machinery inven-tory increases except for the largest cate-gory, those over $350,000, where itdecreases slightly. The importance of hav-ing enough money to pay cash becomesless important as the value of machineryinventory increases to the $249,000 range,then increases in importance for the$250,000 to $350,000 category, and de-creases for the over $350,000 category.Farmers in the 25-to-40 age bracket placeless importance on family persuasion thando farmers in other age brackets. This isthe age when parental influence has wanedand before the influence of children isheeded.

While the above insights are important,of even greater significance is the conclu-sion that these four variables affect theresponses since it suggests that farm man-ager decisions are influenced by variableswhich are treated only indirectly in neo-classical theory. The existence of a man-agement process which allows for the in-clusion of noneconomic factors or criteriain decision making appears to be validat-ed. While a more rigorous statement ofthe model and a broader search for sig-nifcant independent variables is needed,this study indicates that farm managersconsider at least these four factors.

In addition, we are able to reject thenull hypotheses that agricultural represen-tatives and farm machinery dealers per-ceive the management process accurately.These groups, in fact, underestimate theimportance of machinery wearing out, andoverestimate the importance of improved

305

Johnson et al.

Western Journal of Agricultural Economics

features on new models, and income taxconsiderations. These findings may sug-gest the need for these groups to learnmore about their clientele's decision-mak-ing process.

References

Brown, W. J. and R. C. Strayer. "Purchasing FarmMachinery: A Summary of Responses fromFarmers, Agricultural Representatives, and FarmMachinery Dealers." A report submitted to theChairman, Saskatchewan Research Council, Sas-katoon, Saskatchewan, July 1981.

Cromier, P., A. Mitchel, and R. McGriffen. "ProcessIntensive Workshop Manual." Unpublished paper,Winnipeg, March 1982.

Harris, Richard J. A Primer of Multivariate Statis-tics. New York: Academic Press, 1975.

Heck, D. L. "Charts of Some Upper Percentage Pointsof the Distribution of the Largest CharacteristicsRoot." Annals of Mathematical Statistics 31(1960):625-42.

Kepner, C. H. and B. B. Tregoe. The Rational Man-ager. Kepner-Tregoe Inc., Research Road, Prince-ton, New Jersey, 1965.

Morrison, Donald F. Multivariate Statistical Meth-ods. New York: McGraw-Hill, 1967.

Saskatchwan Department of Agriculture. Agricul-tural Statistics-1979. Statistics and Public Infor-mation Branch, Saskatchewan Agriculture, Re-gina, Saskatchwan, 1980.

306

December 1985