Attitudinal Factors Among Respondents

When an attempt is made to summarize a large body of data including many variables by means of relatively few parameters, a number of different statistical methods in the field of multivariate analysis can be applied. In the case at hand, where the aim is to find certain identifiable differences in attitudes towards agriculture and the environment, the question is about an analysis of interde-

pendence in the sense that the set of attitudinal variables taken under closer inspection are assumed to have an equal status. This means that the variables, although expected to be interdependent, cannot be classified as dependent vari-ables and explanatory varivari-ables (Chatfield and Collins 1980, pp. 7-8). The factor analysis (FA)16 applies for this kind of approach. Its idea is to derive new variables called common factors, which are expected to give a better under-standing of the data.

The empirical analysis of respondents' attitudes was based on 22 attitudinal questions included in the questionnaire. They measured respondents' views regarding agriculture, the environment, and sustainable development (question 3-7 in Appendix B, claims from a to v). Ali the questions were presented in the form of claims and the respondents had to agree or disagree with them by expressing their opinion on a five-point scale. These questions were the same in ali the questionnaires, although their placement varied to some extent depending on the type of the questionnaire (see Figure 6.1 in Chapter 6.1). In one half of the questionnaires attitudinal claims were asked after the actual WTP question and in the other half before the actual WTP question.

It appeared that there was significant correlation between most of the attitudinal variables. When testing pairwise Pearson correlations, 18 variables out of 22 variables had at least one correlation coefficient higher than 0.3. This strongly indicated that it was worthwhile to proceed with the application of the factor analysis.

The factor analysis was executed by using the FACTOR procedure that is included in the SAS/STATTm software package. The main option chosen was the principal factor analysis that is based on an orthogonal common factor model. The prior communality estimates were calculated by using squared multiple correlation (SMC)17 (Method I). In addition, two other factor analyses were carried out, a principal factor analysis in which the prior communality estimates were calculated by using maximum absolute correlation of a variable with any other variable (Method II), and a maximum likelihood factor analysis where the prior communality estimates were calculated by applying the SMC (Method III). These two additional analyses were used in order to confirm the results. The rotation method chosen in ali three analyses was the varimax method.

16 The statistical theory behind the factor analysis is not reviewed in this connection, but it can be found in Appendix F.

17 The SMCs may be obtained, for instance, as one minus the reciprosals of the corresponding elements of the inverse factor score matrix. Using the SMCs as communalities limits the analysis to the variance a given variable actually shares with other variables in the data set.

Consequently, variance not shared with the specific set of variables is treated as unique (Bernstein et al. 1987, p. 189).

From the 22 original attitudinal claims 10 were included in the final analysis.

The other variables were excluded because of too low values of pairwise Pearson correlation coefficients or too low estimates of final communality. The rule of thumb was that the exclusion of a variable took place if the highest pairwise correlation coefficient was less than 0.2 or the final communality estimate was less than 0.3. According to Child (1990, pp. 34-35), if the final communality is too low, in the region of 0.3 or less, the increasing existence of unique variance makes the analysis unreliable because it is not possible_ to reject the hypothesis that the major part of unique variance is created by error variance.

In the factor analysis, the most important decision to be made is probably the choice of the number of common factors. Most often, the final choice of the number of common factors is based on some combination of the proportion of sample variance explained, subject matter knowledge, and the general reasona-bleness of the results (Johnson and Wichern 1982, p. 437). In this case, the plausible number of common factors was found to be three, although the appli-cation of different selection criteria did not produce unambiguous results. Occa-sionally, a rule is recommended that only factors which have eigenvalues greater than one should be included. However, this criterion is more suitable for the principal component analysis than for the factor analysis (Child 1990 pp. 37-38). According to this criterion, the maximum number of factors to be extracted would have been from one (Method I) to two (Methods II and III).

Another way to decide the number of factors is to use the so-called Cattell's scree test. The idea is to derive the number of factors from the relations among successive eigenvalues. This inference is usually made graphically by present-ing eigenvalues along the Y-axis and their serial positions along the X-axis. The goal is to separate the overall curve into two functions with the early eigenvalues representing factors that are more important and the later ones representing factors that are less important (Bernstein et al. 1988, p. 174). The Cattell's scree test plots produced by each of the three methods are shown in Appendix E. The conclusion was that the recommended number of factors to be extracted varied from two (Method III) to three (Methods I and II).

It is also possible to use the so-called chi-square test when the maximum-likelihood factor analysis (Method III) is applied. The aim is to determine if the correlation matrix with unity diagonals differs significantly from the identity matrix and if the residual matrix differs significantly from the null matrix because of the extraction of one or more common factors. The problem with the chi-square test is, however, that with large samples a matrix containing trivial residual variance can still differ from a null matrix, resulting in the extraction of trivial factors (Bernstein et al. 1988, pp. 174-175). In this case, the chi-square test indicated that the correct number of factors should be four. When the null hypothesis was that three factors construct a sufficient solution, then the null hypothesis could be rejected at the 0.01% level of significance. Nevertheless,

when the null hypothesis was that four factors construct a sufficient solution, then the risk of rejecting a true null hypothesis increased to almost 15%. Thus, based on the chi-square test, the recommendation was four factors to be ex-tracted.

Although the criteria guiding the factor selection process proved not to be unambiguously interpretable, the rotated factor pattems were very similar in each case. Based on general knowledge about the subject matter and on certain aspects of the applied selection criteria, especially Cattell's scree test, a three-factor solution was considered to have the most desirable features. It offered an illuminating explanation that did not actually alter when a fourth factor was included. The rotated factor pattem produced by Method I is presented in Table 7.1

When the results presented in Table 7.1 are interpreted, we can see that the percentage variance is quite low. This value tells how large a portion of the total variance is explained by the common factors. In this case, the portion is approxi-mately 43%. Thus, about 57% of the total variance is due to specific or error variance. However, the inclusion of a fourth factor would have increased the common variance only by 0.3%. This fact also supports the selection of the three-factor model. However, the main purpose of Table 7.1 is to give informa-tion that makes it possible to develop meaningful descripinforma-tions for each factor.

Table 7.1. Variables, Factor Loadings'8, Communalities, Eigenvalues, and Per-centage Variance in the Varimax-Rotated Principal Factor Solution of Three Factors (Method I).

Variable Factor 1 Factor 2 Factor 3 112

X136 0.683 0.282 0.258 0.613

X137 0.593 0.182 0.224 0.435

X135 0.526 0.384 0.269 0.497

X138 0.509 0.209 0.373 0.441

X144 0.207 0.672 0.009 0.495

X141 0.313 0.501 0.166 0.377

X140 0.272 0.478 0.182 0.336

X145 -0.068 -0.608 0.114 0.387

X133 0.245 0.071 0.577 0.398

X134 0.183 -0.053 0.573 0.365

Eigenvalues 1.667 1.613 1.063 4.343

Percentage variance 16.672 16.128 10.629 43.429

18 Factor loadings higher than 0.3 are in bold. The criteria for the detection of significant or salient factor loadings are somewhat vague. However, a rule of thumb, very widely used by factor analysts, is that factor loadings having values of ±0.3 or greater are usually regarded as significant, on condition that the sample size is greater than 100 (Child 1990, p. 39).

The following variables received significantly high positive loadings on Fac-tor 1: X136, X137, X135, X138, and X141.

X136: The use of fertilizers and pesticides is at too high a level in Finnish agriculture

X137: The intensification of Finnish agriculture deteriorates the environ-mental quality and food safety

X135: Environmental problems caused by agriculture are already signifi-cant

X138: Conventional agriculture should quickly be developed into the di-rection of organic farming

X141: Agriculture has to carry its fair share of the environmental taxes The following variables received high positive loadings on Factor 2: X144, X141, and X140. In addition, variable X145 had a high negative loading and variable X135 received a significantly high positive loading on Factor 2.

X144: Agricultural subsidies financed by taxpayers must be cut down if agriculture is not able to pioduce foodstuffs at competitive prices X141: Agriculture has to carry its fair share of the environmental taxes X140: Conventional agricultural subsidies can be cut down if the

cone-sponding amount of money will be used to promote environmental investments and environmentally related subsidies in agriculture X135: Environmental problems caused by agriculture are already

signifi-cant

X145: It is a right thing to allocate tax money for the maintenance of agriculture because the viability of the countryside and the pleas-antness of the environment depend on agricultural activities The following variables received high positive loadings on Factor 3: X134, and X133. In addition, variable X138 received a significantly high positive loading on Factor 3:

X134: The present generation must take better care of the environment that will be left to the coming generations

X133: Environmental conservation should have greater_emphasis, even at the expense of economic growth

X138: Conventional agriculture should quickly be developed into the di-rection of organic farming

The interpretation of the three factors is the following:

Factor 1: This factor represents an attitude the leading argument of which is that adverse environmental impacts of conventional farming prac-tices constitute the most severe problem in Finnish agriculture. The

most preferable solution is to develop current farming practices into the direction of organic farming, and in this process it is even acceptable to use economic instruments.

Factor 2: This factor can be seen to represent an attitude the core element of which is dislike towards conventional agriculture on the grounds of additional tax burden caused by agricultural subsidies. The empha-sis is not on adverse environmental effects of agriculture, although environmentally beneficial farming practices are not opposed. How-ever, a high positive loading of variable X140 and a high negative loading of variable X145 seem to contradict to some extent. The valid interpretation is probably that the promotion of environmen-tally beneficial farming practices is acceptable only if the govem-ment can simultaneously guarantee that the total amount of tax money allocated to agricultural subsidies decreases. Nevertheless, adverse environmental effects of farming are more like an excuse to criticize agriculture than a source of true concem.

Factor 3: This factor expresses an attitude, which emphasizes the importance of sustainable development in agriculture. Environmental issues have a high ranking. The major source of concem is the damage that might already have been caused to sustainable development and future generations because of the undiscriminating admiration of economic growth. As a part of this general framework, organic farming is seen to he a preferred altemative when the future of agricultural production is concemed.

The factors differ in respect of both economic and environmental issues.

Factor 1 and Factor 2 represent views that are clearly critical towards conven-tional agriculture, although for different reasons. 'When it comes to Factor 2, it seems that the resistance culminates on the grounds that are most probably derived from standpoints related to income distribution and income transfers.

The bottom line is that the society should subsidize farmers only if this leads to detectable efficiency gains. This view mainly concentrates on economic aspects and is quite insensitive in terms of environmental concems. Factor 1 conveys a more pragmatic approach. The use of taxpayers' money to support agriculture is not undesirable as such, but it becomes undesirable when the outcome is nega-tive in the form of adverse environmental effects. However, the expression of environmental concem is not as genuine and fimdamental as it is in the case of Factor 3. It is more like a signal of discontent with an investment the profits of which have not been as large as expected. Thus, Factor 3 is the only factor that

represents a true environmental concem, while agricultural subsidies and their effect on social welfare is ali but ignored.

7.2. Further Division of Respondents into Clusters Based on Attitudes