Parents' Judgments and Students' Self-Judgments ... - Hogrefe eContent

6 downloads 0 Views 207KB Size Report
C.Cliffordson:Parents' JudgmentsandStudents' Self-Judgments ofEmpathy. EJPA17 (1),© ...... ment, “I experience distress and discomfort in emergen-.
European Journal of Psychological Assessment, Vol. 17, Issue 1, pp. 36–47 C. Cliffordson: Parents’ Judgments EJPAand 17 (1), Students’ © 2001Self-Judgments Hogrefe & Huber of Emp Publishers athy

Parents’ Judgments and Students’ Self-Judgments of Empathy The Structure of Empathy and Agreement of Judgments Based on the Interpersonal Reactivity Index (IRI)* Christina Cliffordson Department of Education, University of Göteborg, and Trollhättan/Uddevalla University, Sweden Keywords: Empathy, Interpersonal Reactivity Index, IRI, self/other judgment, structural equation modeling Summary: A sample of 221 adolescents and their parents were asked to complete the Interpersonal Reactivity Index (IRI; Davis, 1980), which is comprised of four subscales: empathic concern, perspective taking, fantasy, and personal distress; and a global-item measure containing four items, one for each subscale. The factorial structure that emerged in a previous study (Cliffordson, 2000) involving the students was tested on ratings provided by their parents. The results from the IRI scale were also compared to results from the global-item measure. Structural equation modeling was used to examine the structure of empathy and the agreement of self/other judgments. The results support the conclusions from the previous study that the concept of empathy can be considered to be identical to empathic concern, which also explains a great deal of perspective taking and fantasy. The agreement between the students’ and their parents’ judgments was substantial, and there are several reasons to believe that the interjudge agreement obtained is accurate.

The Structure of Empathy Empathy is an important concept in areas of personality research and research regarding interpersonal behavior. However, there remains considerable confusion concerning the definition and measurement of empathy. The main conflict concerns the distinction between empathy as a cognitive and as an emotional phenomenon. Another discussion concerns the dimensionality of the construct, that is, whether it should be seen as a unidimensional construct or whether a multidimensional approach in studying empathy is needed. The most widely used measures of empathy vary both regarding the nature of the construct and its internal structure (cf. Interpersonal Reactivity Index (IRI); Davis, 1980; The Hogan Empathy Scale (HES); Hogan, 1969; Mehrabian and Epstein

Questionnaire Measure of Emotional Empathy (QMEE); Mehrabian & Epstein, 1972). While the QMEE was developed as a measure of emotional empathy, reflecting the tendency to respond emotionally to the experiences of others, the HES may be considered as a more cognitive measure developed from the definition that empathy is “. . . the intellectual or imaginative apprehension of another’s condition or state of mind without actually experiencing that person’s feelings” (Hogan, 1969, p. 308). In contrast to these two viewpoints, Davis argues (1980, 1983, 1996) that, rather than treating empathy as a single unipolar concept, empathy may best be considered a set of constructs. Davis developed a self-report measure, the IRI scale (1980), based on a multidimensional approach. The instrument aims at providing measures of dispositional tendencies

* The original data upon which this paper is based are available at http://www.hhpub.com/journals/ejpa EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

in four areas, labeled perspective taking (PT), empathic concern (EC), fantasy (FS), and personal distress (PD). The set of constructs is related in that they all concern responsivity to others but are also clearly discriminable from each other: – The EC scale aims to assess the affective outcomes, the tendency to experience other-oriented feelings and the response to distress in others with the reactive response of sympathy and compassion. – The PT scale aims to measure the process of role taking, the tendency to adopt the psychological points of view of others. – The PD scale also demonstrates an affective outcome, and is designed to tap ones’ own feelings of personal unease and discomfort in reaction to the emotions of others. – Finally, the FS scale aims to measure the tendency to transpose oneself into feelings and actions of fictitious characters. Davis (1983) assessed the validity of the IRI scale by investigating its relationships with measures of social functioning, self-esteem, emotionality, and sensitivity to others, as well as with previously developed unidimensional empathy measures. The pattern of relationships predicted by Davis was confirmed. The investigation of the relationships showed that three of the subscales were significantly and positively related (FS – PT = .10, EC – PT and EC – FS = .33, respectively). The correlations between each of the IRI subscales and the extant empathy measures (QMEE and HES) displayed significant positive relations, except for the relation between PD and HES. As was expected, the HES achieved the greatest degree of correlation with the cognitive PT scale (.40), and the QMEE achieved the greatest degree of correlation with the emotional EC scale (.60) and the FS scale (.52). A similar pattern of intercorrelations of the IRI subscales, and relations between the subscales and the empathy measures, has been discovered by Johnson, Cheek, and Smither (1983), Wise and Cramer (1988), and Riggio, Tucker, and Coffaro (1989). Thus, the measures mentioned above cover different aspects of the global concept of empathy and, with the exception of the PD subscale, they all are positively related. When different measures show intercorrelations, there are reasons to expect that they reflect the same hypothetical construct (Snow, Corno, & Jackson III, 1996). Of the most commonly used empathy measures the IRI scale seems to be the most comprehensive one, which is why an analysis of the structure of the IRI scale might give a valuable contribution to the understanding of the structure of empathy. In a previous study (Cliffordson, 2000) involving 221 adolescents, the structure of empathy was investigated by examining the IRI scale

37

(Davis, 1980). Confirmatory factor analysis (CFA) was performed to assess the dimensionality of the scale, the relationships among the subscales, and to compare the results from IRI with results from a self-report measure consisting of four global items, one for each of the IRI subscales (GI measure). An important guide to assessing the appropriateness of items of an existing scale is the definition of the construct (Pedhazur & Pedhazur, 1991). The use of one global item for each of the four subscales, based on Davis’ (1980, 1983, 1996) definitions of the constructs and descriptions of what each of the subscales intends to measure, implies that the content of the four subconstructs and the concept of empathy is concretized by the subjects responding to the items. The validity of the dimensionality of the IRI scale was in that way investigated by comparing the results as regards the content of empathy, as arrived at by Davis, with those yielded by the subjects who answered the GI measure. The results from the Cliffordson (2000) study support Davis (1980, 1983, 1996) in that the IRI subscales tap aspects of a global concept of empathy. The analysis showed, however, in contrast to Davis’ four separate aspects, a single general empathic dimension that was close or identical to EC and that also explains a great deal of PT and FS. These results agree with those obtained by Wise and Cramer (1988). Another orthogonal personal distress dimension was also found that was almost equal to PD. The content of PD, conceptualized by the respondents appears, however, to be different when compared with PD as concretized by Davis (1980). Thus, the validity of PD and therefore also the personal distress dimension, can be questioned. Because these results seem to contradict previous research (Davis, 1980, 1983, 1996) on the structure of the IRI scale and the concept of empathy, and because the results are based on only one sample, there are reasons for further investigations. One purpose of the present study is to investigate the internal structure of empathy by examining the validity of the IRI scale adapted to the judgments of others. The factorial structure that emerged from the previous study involving the students will be tested on judgments provided by their parents. The results from the IRI scale will also be compared with results from the GI measure.

Agreement of Judgments A frequently used criterion for accuracy is interjudge agreement, which is a matter of convergent validity (Funder, 1987; Funder & Colvin, 1988). Funder and Colvin (1997) assert that it appears that self/other agreement in personality judgments is a fairly robust phenomenon, and that the only reasons for lack of agreement appears to be when the acquaintanceship is weak, when EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

38

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

unreliable rating scales are used, or when a strong selfpresentational pressure is present. Two different methods to assess self/other agreement in personality judgments are most frequently used (Funder & Colvin, 1997). One implies the analysis of mean differences, while the other involves the analysis of correlations. In areas such as assessment of traits, it is implausible to assume that the conceptual variables are measured anywhere near perfectly (Maruyama, 1997). That is, judgments of traits like empathy always include error components. When measurement error is present, the results emerging from the methods described above, and particularly so the one based on correlations, become biased. Thus the results cannot be trusted to accurately reflect the agreement between the judgments of others and self-judgments. The mean difference method, however, is not necessarily burdened with the problem of random errors of measurement, but with the problem of variance unique to the variable. However, when SEM is used, the analyses are based on correlations and mean differences between latent variables, which reflect the common, reliable parts of the manifest variables (Gustafsson & Stahl, 1997). Another purpose of this study is to examine self/other agreement in judgments of empathy by performing SEM in order to assess the agreement between students’ selfdescription of empathy, and the judgments provided by their parents. There are reasons to believe that students’ self-descriptions are more favorable than descriptions provided by their parents. Several studies have shown that people manifest more positive opinions of themselves than others have of them (cf. Kunda, 1987). However, the selfenhancement effect is much stronger when comparing self-ratings to ratings by strangers than when the comparison is with ratings by close acquaintances (Funder & Colvin, 1988). In addition, the IRI scale taps dispositional tendencies, which are more or less observable by others. Previous research (e. g., Funder, 1980) shows that people give higher ratings to themselves on internal (invisible) traits than do their acquaintances, whereas acquaintances tend to give higher ratings to people on external (visible) traits.

Method Measures The two empathy measures, the IRI scale (Davis, 1980) and the GI measure, were the same for the students and their parents, with the exception that the measures utilized for the parents were adapted to their points of view. EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

In responding to the measures, the subjects were asked to indicate the degree to which the items described themselves and their son/daughter, respectively, by choosing the appropriate point on a five-point Likert scale. The Interpersonal Reactivity Index (IRI) The IRI scale (Davis, 1980) contains four seven-item subscales. A sample of items from the four subscales adapted to the parents’ point of view was: – “I would describe my son/daughter as a pretty softhearted person.” (EC); – “My son/daughter believes that there are two sides to every question and tries to look at them both.” (PT); – “In emergency situations, my son/daughter feels apprehensive and ill-at-ease.” (PD); – “My son/daughter daydreams and fantasises, with some regularity, about things that might happen to him/her.” (FS). The properties of the measure reported by Cliffordson (2000) indicate that the Swedish version of the IRI scale (Kulich & Bengtsson, in preparation) is comparable to the original version (Davis, 1980). The intercorrelations of the four scales were almost equal to the corresponding results presented by Davis (1983). The alpha reliability of the four subscales adapted to judgments of others is acceptable (EC = .77, PT = .77, FS = .80, and PD = .74) and similar to both the corresponding result from the previous study (ranging from .71 to .80) and the corresponding result presented by Davis (ranging from .71 to .77). The Global-Items Measure (GI) The author, on the basis of the IRI scale, constructed the four-item measure consisting of one item for each dimension. The wordings of the items from the parents’ point of view were: – “My son/daughter finds it easy to experience feelings of sympathy and compassion for unfortunate others.” (EC); – “My son/daughter finds it easy to see things from other peoples’ perspectives.” (PT); – “My son/daughter experiences distress and discomfort in emergencies or when other people are in extreme distress.” (PD); – “My son/daughter is imaginative and finds it easy to transpose him/herself into fictional situations.” (FS).

Subjects and Procedures The subjects were 221 11th-grade students, at an upper secondary school in a town in the western part of Sweden, and their parents. The data collection took place in

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

May 1998. After the students had been administered the GI measure and the IRI scale personally by the author, the parents were asked to judge their adolescents on parallel measures. Instructions to the parents for responding to the measures were given in writing (letter). Both students and parents were informed that participation was voluntary, and that a guarantee of anonymity was given. The students returned the completed measures.

Methods of Analysis The number of internal missing values was negligible and distributed over the items and the participants, with the exception of two parents, one who answered only one of the two measures and another who left as much as ten items unanswered on the IRI scale. These two participants were for this reason excluded from the analyses. The internal missing value (30) was replaced with the mean for each item. Finally, out of the 221 cases, each consisting of data from both the students and their parents, 137 cases had complete data and 84 cases had missing data on all parents’ variables. The problem with the missing parents’ scores was handled by utilizing the modeling procedure based on theory and methodology for improving validity in studies with missing data, as described by Arbuckle (1996) and Muthén, Kaplan, and Hollis (1987). In the present study, the data consists of two subsamples: one subsample with complete data from both students and their parents (n = 137), and another subsample with incomplete data or missing data on all of the parents’ variables (n = 84). The basic principle of the modeling procedure is that the subsample with missing data is not necessarily from the same sample as the subsample with complete data. In this approach, the information in the available data is used to predict the missing values. In order to investigate the structure of empathy, CFA models previously fitted to the data from the students (Cliffordson, 2000), were now fitted to corresponding data from the parents. SEM models defining identical measurement models for the students and their parents, including equality constraints for corresponding estimates over the measurement models, were specified and tested. In order to investigate the self/other agreement between students and parents, covariances for corresponding factors over the measurement models were examined. The models were also tested with regard to mean differences by comparing the means between the corresponding factors. These models were fitted to the covariance matrix for the part of the data that was complete for both students and parents (n = 137). The tools used for factor analyses were Lisrel (Jöreskog & Sörbom, 1996) and Amos (Arbuckle, 1997) with the STREAMS environment for specifying, esti-

39

mating, and evaluating the models (Gustafsson & Stahl, 1998). As measures of model fit the χ2 goodness-of-fit test and the root mean square error of approximation (RMSEA) were used. The RMSEA, which measures the amount of discrepancy between the model and data in the population, taking the number of estimated parameters into account, is strongly recommended as a tool in the evaluation of model fit (Bollen & Long, 1993).

Results The descriptive statistics suggest that one item should be excluded from the analyses because of an unacceptably skewed distribution. The means and standard deviations for the items are shown in Table 1. In five out of 32 items, the means were significantly higher for the students with complete data in comparison to the students with incomplete data. The comparisons, regarding the means, between parents and students showed that in 20 out of 32 items the differences were significant. Of those, some were higher for the students and others higher for the parents.

The Structure of Empathy The Hypothesized Four-Factor Model The first step was to investigate an hypothesized fourfactor model with one latent variable for each of the four subscales (Ec, Pt, Fs and Pd). The model includes covariances between the four latent variables (Model A, Figure 1). The fit indices for the model applied to the students’ data (χ2 (344, N = 221) = 700.99, p < .00; RMSEA = .069) indicates that the model is reasonably good, but the results also indicate that there is room for improvement of the model. When applied to the parents’data the model shows, however, a poorer fit (χ2 (318, N = 137) = 570.59, p < .00; RMSEA = .076). The hypothesized four-factor model was used for further modeling. The Eight- and Ten-Factor Models In the previous study (Cliffordson, 2000), several alternative models were tested. One model including correlations among some of the error variances of the manifest variables fitted well. This is one way to improve the fit of a model. Another approach is to specify specific factors for the correlated items. The model, which was used for further modeling, was a model with eight factors, one for EC, one for PT and three factors each for PD and FS. This model fitted equally well. The relationships among EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

40

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

Table 1. Descriptive statistics for the parents’ judgments and the students’ self-judgments. Item

Empathic Concern – Global 14. Peoples’ misfortunes disturb me. 18. Feel pity, treated unfairly. 20. Touched by happenings. 22. Describe myself as soft-hearted. 2. Feel concerned, less fortunate. 4. Feel sorry, having problems. 9. Feel protective, taken advantage. Fantasy – Global 12. book/movie, get involved in. 16. Play/movie, feel as characters. 1. Daydream and fantasize. 23. Movie, put myself in character. 26. Story/novel, feel happening me. 5. Novel, feel like characters. 7. Movie/play, get caught up in it. Personal Distress – Global 10. Emotional situation, helpless. 13. Someone get hurt, remain calm. 17. Tense situations scares me. 19. Dealing emergencies, effective. 24. Emergency, lose control. 27. Emergency, “go to pieces”. 6. Emergency, feel apprehensive. Perspective Taking – Global 11. Look from others’ perspective. 15. Listening to others’ argument. 21. Look at two sides of a question. 25. Put myself “in his shoes” 28. Feel in place of others. 3. See from others’ point of view. 8. See others’ side before decision.

Mean Complete Data Parents Students 4.40 4.06 4.46 3.75 4.74*3 3.84 3.87 4.10 3.92 3.34 2.50 2.60 3.09 3.25 3.53 2.79 3.93 3.18 2.70 2.56 2.32 2.00 2.56 3.31 3.96 3.72 2.94 3.64 3.04 3.59 3.65 3.80

4.04*1 3.91 4.29*1 3.76 4.25*1 3.53*1 3.54*1 4.07 3.90 3.77*1 2.83*1 4.04*1 3.39*1 3.62*1 3.90*1 3.47*1 3.95 3.38 2.69 2.55 2.53*1 2.31*1 3.16*1 3.31 3.78*1 3.85 3.39*1 3.73 2.67*1 3.31*1 3.42*1 3.76

Incomplete Data Students

SD Complete Data Parents Students

Incomplete Data Students

3.83 3.83 4.05*2 3.60 4.07 3.44 3.62 4.00 3.58*2 3.69 2.52 3.77 3.01*2 3.37 3.63 3.29 3.61*2 3.13*2 2.59 2.44 2.51 2.35 2.92 3.10 3.65 3.74 3.31 3.58 2.53 3.28 3.51 3.60

.79 .92 .68 .83 .52 .96 1.06 .79 .99 1.17 1.07 1.13 1.08 .93 1.06 1.20 .95 1.00 1.05 1.14 .76 .89 1.05 1.15 .89 .77 1.21 .81 .87 .96 1.05 .86

.89 .92 .94 .84 .77 .95 .97 .70 .88 1.23 1.24 1.10 1.17 1.11 1.25 1.18 .99 .82 .98 1.00 .87 .88 1.06 1.13 .81 .82 1.21 1.02 1.08 1.01 .93 .84

.93 .91 .73 .81 .68 .87 1.08 .74 .88 1.19 1.23 .93 1.04 1.04 .93 1.25 .96 .88 1.14 .95 .77 .86 1.09 1.09 .79 .84 1.19 .93 1.09 .98 .92 .94

Note. Subsample with complete data, n = 137; subsample with incomplete data, n = 84. *1 = mean differences between the parents and the students from the subsample with complete data are significant beyond the .05 level, *2 = mean differences between the students from the subsample with complete data and the students from the subsample with incomplete data are significant beyond the .05 level. (sig. are two-tailed). *3 = is excluded from further analyses.

the three Pd and the three Fs factors respectively were, however, comparatively strong (ranging from .44 to .75), which provide support for the existence of one common dimension for both the PD and the FS dimension. Two second-order factors were introduced in the eight-factor model, each of which related to the three Pd and the three Fs factors respectively. This model fitted almost as well as the eight-factor model. These results did not completely confirm Davis’ (1983) assertion that the IRI subscales measure four discernibly different dimensions, even though the main pattern of results was expected from the results reported by Davis. The Higher-Order Model with One Global Factor The final and most appropriate model for the IRI scale was a higher-order model (Model B) with one global EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

factor (Emp) related to the four factors, of which each represents one of the four dimensions (Ec, Pt, Fs and Pd). A model for both the students and their parents, including covariances for the corresponding factors (Emp, Pt&, Pd&, Fs&, Fs1&, Fs2&, Fs3&, Pd1&, Pd2& and Pd3&) fits well (χ2 (1394, N = 137) = 2067.56, p < .00; RMSEA = .060). Standardized factor loadings of the four factors on Emp display a perfect relationship between Emp and Ec (1.0) for both students and parents, while the relationships between Emp and Pd are much weaker and different between students (–.03) and parents (.22). The relations between Emp and the other two factors are, however, similar between the two samples and greater (ranging from .47 to .63). A more restrictive model, with the Ec residual variances fixed to zero fitted equally as well as previous model (∆ χ2 (2, N = 137) =

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

41

was tested by releasing the remaining category of parameters from the invariance restriction. The results obtained from these analyses show that the test of equality of factor loadings is significant (∆ χ2 (29, N = 137) = 63.08, p < .00) as is the test of equality of variances in latent variables and error variances (∆ χ2 (38, N = 137) = 88.38, p < .00). Furthermore, in order to find out which of the single parameters are significantly different, the constrained parameters are released from the invariance restriction one by one. Significant differences between students and parents are found in five out of 40 factor loadings (PT21 on Pd3, PD27 on Ec, PD13 on Pd1, PD27 on Pd1 and Pd3 on Pd), and in six out of 38 error variances (PT11&, PT21&, PT25&, FS5&, FS16& and PD24&). When the significantly different parameters were released from the invariance restriction the model fit improved. The difference between the partly constrained model and the model in which all parameters are allowed to vary is nonsignificant (∆ χ2 (56, N = 137) = 66.49, p < .16). The results from the higher-order model analyses confirm the results from the previous study (Cliffordson, 2000), that empathy is organized hierarchically with one general dimension at the apex. The results from the present study imply that under the four scales there is one dimension or theoretical construct (Emp) which accounts for 100% of the variance in Ec, between 22%–40% of the variance in Pt and Fs, and between none and 5% of the variance in Pd. The significant differences in factor loadings are restricted to the PD dimension (IRI), which support previous conclusion that the PD dimension is uncertain. Models Including the GI Measure

Figure 1. The hypothesized four-factor model with covariances among the four factors (Model A).

.27, p < .87), which indicates that the relation between Emp and Ec is unity for both the students’ and the parents’ data. The same model including equality constraints for the corresponding factor loadings and variances over the two measurement models was also tested. A test of equality of the two measurement models is obtained if the fit of the model that imposes no constraint of equality is compared with the model that imposes full constraints over the two models. This overall equality test shows that the difference is significant (∆ χ2 (67, N = 137) = 144.09, p < .00). To find out in what parts of the model differences are located, the equality of each category of parameters

The next step was to examine the model (Model C) including the four global items, each focusing one of the four subconstructs. The test of equality between students and parents for this model is nonsignificant (∆ χ2 (8, N = 137) = 8.93, p < .35), which means there are no significant differences in factor loadings, variances in latent variables, or error variances. In order to validate the results which emerged from the higher-order model with one global factor, a model with two global factors was specified and tested (model D, Figure 2). The higher-order model (Model B) was extended with the model with one global factor related to the four manifest variables (Model C). As is clear from Figure 2, the model contains covariances relating to the residual variables for the corresponding components of empathy in the two sets of measures, with the exception of the EC (GI) and the Ec (IRI) residual since the Ec residual in the IRI model lacks EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

42

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

Figure 2. The higher-order model with two global factors (Model D), including estimated covariances and factor loadings for students and parents (students’/parents’ estimate, respectively; shadowed numbers indicate significant differences beyond the .05 level).

variance. The model also includes a covariance among the two global factors. A model for both the students and their parents fits well (χ2 (1841, N = 137) = 2739.26, p < .00; RMSEA = .060). The significantly different parameters that were EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

found between the students and their parents, are the same in this model as in the previously presented model (Model B), with the exception of a marginally significant error variance of one manifest variable (PD17&). On the contrary, no significant differences were found in the

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

43

Table 2. Correlations between latent variables for the students’ and their parents’ judgments. All parameters are free in the models. Latent variable Emp 1 Emp 2 Ec Ec& Pt Pt& Fs Fs& Fs1& Fs2& Fs3& Pd Pd& Pd1& Pd2& Pd3&

Four-factor model Complete data

Correlations Higher-order model (D) Incomplete data Complete data ncomplete data

.39**

.40**

.21

.20

.45**

.23**

.32** .43**

.33** .39**

.66

.79

.15

.15

.44** .01 .59** .08

.43** .00 .57** .11

.47**

.24** .16 .09 .26 –.03

.12 .08 .27 .13

Note. Complete data, n = 137; incomplete data, n = 221. ** = significant beyond the .05 level.

variances of the latent variables, the variances of residuals in latent variables or the covariances. When the significantly different factor loadings (five out of 44, see Figure 2) and error variances (seven out of 43) were released from the invariance restriction, the differences are nonsignificant (∆ χ2 (67, N = 137) = 82.57, p < .09). As is shown in Figure 2, the correlations between the two global factors are strong for both students and parents. The pattern of factor loadings is similar between students and parents, with the exception of the factor loadings in the PD dimension. In summary, the findings from the examination of the IRI scale show that both the standardized factor loadings on the two global factors, and the strong correlations among the two global factors are similar and not significantly different between the two samples. That is, the results from previous study (Cliffordson, 2000) are replicated, which supports the conclusions that there is one dimension, which appears to be a general empathic factor, close or identical to EC and that this dimension also explains a great deal of PT and FS. The factor loadings also show that the responses on the PD items are significantly different between students and parents. The content of PD from the parents’ responses on the seven-item subscale (IRI) appears to be more similar to both the students’ and the parents’ responses on the global PD item (GI). That is, the parents’ responses on the PD subscale appear to be more similar to all the subjects’ conceptualizations from the definition of the PD aspect by Davis (1980, 1983, 1996). While the standardized factor loading of Pd on Emp1 for the students is negative and nonsignificant (–.03) the corresponding factor loading for the parents is positive and significant (.23). These results support the

previous conclusion that the PD, concretized by Davis through the seven-item scale, is doubtful. The Impact of Missing data The results from the missing-data analyses for the higher-order model including two global factors (Model D) show no differences with respect to the covariances, whereas variances and factor loadings appear to be different. While the test of equality for the complete data model shows significant differences in variances (∆χ2 (43, N = 137) = 117.00, p < .00), the test of equality for the incomplete data model is nonsignificant (∆ χ2 (43, N = 221) = 27.60, p < .97). The equality test for the model where the significant parameters are released from the invariance restriction is, however, still significant for the incomplete data model (∆χ2 (67, N = 221) = 156.00, p < .00). The investigation in order to find out which of the factor loadings are significantly different between students and parents shows that almost all factor loadings for the IRI scale are significantly different. The structure of the factor loadings (Incomplete data, Model D, students/parents: Emp1 on: Ec = 1.0/1.0, Pt = .52/.57, Fs = .48/.40, Pd = .02/.22; Emp2 on: EC = .82/.74, PT = .54/.61, FS = .25/.06, PD = 57/57, cf. corresponding estimates Figure 2) and the pattern of differences between factor loadings for students and parents obtained from the two kinds of analyses are quite similar. The correlations among the two global factors are also similar (.80/.91, cf. Figure 2). Thus, in spite of the differences in factor loadings, the conclusion from the missing-data analyses is that the main results regarding the internal structure from the complete data models are valid for the study. EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

44

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

relation of the Ec residuals from the higher-order model is nonsignificant.

Agreement of Judgments Correlations Between Latent Variables For purposes of examining judgmental agreement, correlations between the two sets of judgments are assessed. The analyses are based on correlations between latent variables for the students’ and their parents’ judgments. The main results are the significant correlations between the global empathic factor, as measured by both the IRI scale and the GI measure. The correlation is, however, somewhat smaller for the Emp1 factor (IRI) than for the Emp2 factor (GI). One explanation is the differences in responses on the PD items described above. The result from the analysis based on the four-factor model, which is comparable to the summed scores for each subscale, shows no significant correlation for Pt, while Fs shows the greatest correlation. The estimates of correlations from the four-factor analysis typically are higher than estimates based on Pearson correlation coefficients for summed scores for each subscale (EC = .39 vs. .27; PT = .21 vs. .20; FS = .45 vs. .38; PD = .23 vs. .19). One reason for this is that the factor analysis sorts out the error variance and solves the problem of attenuation due to errors of measurement. Because the global empathic factor is almost equal to Ec and accounts for a greater or smaller part of the correlations on the remaining three factors, the higher-order model yields a different pattern of correlation estimates. That is, a great deal of the correlation found for the Emp factor is due to the correlation of the Ec subscale and consequently the cor-

Mean Differences in Latent Variables The second way to investigate self/other congruence in judgment is to compare the means of students’ self-description and judgments by their parents. As shown in Table 3, the global empathic factor has a significantly higher level for the parents. The means on Fs and Pd are, on the contrary, higher for the students, whereas no significant difference is visible on Pt. The findings from the mean comparisons in this study are in accordance with previous results (e. g., Funder, 1980) that people tend to give themselves higher ratings on traits that are relatively internal or not outwardly observable than do their acquaintances. Fantasies, daydreams and worries also are traits mentioned by Funder and Colvin. (1997, p. 627–628) as, “. . . more observable by people who possess them than by acquaintances, because fantasizing, worrying, and other internal activities are only accessible to acquaintances to the extent that the persons who perform them let others in on the secret.” The missing-data analyses regarding correlations show that almost all estimates are approximately the same in both kinds of analyses. The pattern of mean differences based on complete data in comparison to those obtained from the incomplete data, show that the profile of the results are consistent. Hence, the conclusion is that the results presented above hold for the study when missing data is taken into account (see Tables 2 and 3).

Table 3. Mean differences in latent variables for the students’ and their parents’ judgments. All parameters are invariant in the models. Latent variable Emp 1 Emp 2 Ec Ec& Pt Pt& Fs Fs& Fs1& Fs2& Fs3& Pd Pd& Pd1& Pd2& Pd3&

Four-factor model Complete data

Mean differences Higher-order model (D) Incomplete data Complete data Incomplete data

–.19**

–24**

–.05

–.09

.47**

.22**

–.29** –.29**

–.23** –.38**

–.19**

–.23**

–.04

–.07

.32** .51** .61** .52** .32**

.39** .45** .51** .23**

.23** .18 .07 .30**

.14 .10 .01 .29**

.18**

Note. Complete data, n = 137; incomplete data, n = 221. Positive numbers indicate a higher level for the student and negative numbers a higher level for the parents. ** = significant beyond the .05 level. EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

Discussion One purpose of this study was to investigate the structure of empathy by examining the validity of the IRI scale (Davis, 1980) adapted to the judgments of others. The results from a previous study (Cliffordson, 2000) regarding students were compared with the judgments provided by their parents. Another purpose was to assess self/other agreement of judgments of empathy by computing correlations and mean differences between students’ self-description and judgments by their parents.

The Structure of Empathy The results from the present research support Davis’ (1980, 1983, 1996) findings that the IRI subscales tap aspects of the global concept of empathy. The results also support the conclusion from the previous study (Cliffordson, 2000) that empathy can be best understood in terms of one general empathic dimension close or identical to EC, and that this dimension also explains a great deal of at least two of the remaining three aspects (i. e., PT and FS). According to Davis the four aspects of the concept of empathy are clearly discriminable from each other and nonhierarchically organized, which implies a distinction between emotional and cognitive components. While EC and PD are described as emotional components and PT as a purely cognitive component, FS is described as being more indeterminable. The conclusions from the previous and present studies challenge Davis’ (1980, 1983, 1996) position, as well as other descriptions of empathy as being either a cognitive or an emotional concept (e. g., Hogan, 1969; Mehrabian & Epstein, 1972). The results indicate that the concept of empathy can neither be seen as consisting of either a cognitive or an emotional component nor as a set of cognitive and emotional constructs clearly discriminable from each other, but rather as one dimension containing both components. The fact that the general empathic dimension is more or less identical to EC and underlies a great deal of both PT and FS indicates that empathy includes both emotional and cognitive aspects. It also indicates that empathy constitutes an integrated entirety with its main emphasis on emotional reactivity to others contributed to by cognitive processes. The interpretation of the general empathic dimension can be compared to Davis’ assertion (1996, p. 217–218) that “. . . it becomes apparent that there are reliable associations between cognitive processes . . . and cognitive outcomes . . ., and that these outcomes contribute meaningfully to the experience of particular affective outcomes . . . while affective outcomes are the most im-

45

portant proximal influence on behavior, cognitive processes and outcomes also contribute directly and substantially to these affective reactions.” The validity of the general empathic dimension is supported by the strong correlation (.69 and .91) between the two global factors that emerged from the model including both the measures (IRI and GI, Model D). The IRI scale is constructed as four independent measures and used for its four aspects, and not as a total score because each factor has a unique psychological meaning and may yield different results in different types of research. The conclusion from the present study, that empathy can be considered as being identical to EC and also explains a great deal of PT and FS as well, indicates that these three subscales, summed up into a total score, might serve as a measure of a general empathic disposition. The alpha reliability for such a scale is somewhat better (.83) than it is for the subscales (ranging from .71 to .80). When the models previously fitted to the data from students (Cliffordson, 2000) also were fitted to data from their parents substantial differences are found for the PD aspect only. In contrast to the students’ responses, it appears that the content of PD from the parents’ responses on the PD items (IRI) are more like the conceptualizations made by both students and parents, from the definition of the aspect by Davis (GI). When the subjects made the conceptualization the relationship between the general empathic dimension and PD appears to be substantial. Thus, the subjects tend to respond to the statement, “I experience distress and discomfort in emergencies or when other people are in extreme distress” with other-oriented feelings like Davis’ EC aspect (sympathy and concern for unfortunate others) or with responses similar to the EC factor described by Litvack-Miller et al. (1997), namely, with feelings of concern or upset about the plight of others. The theoretical understanding of PD as a potentially important aspect of the reactions to others as described by Davis (1996) appears reasonable. It does seem, however, that there are difficulties in concretizing and measuring the PD aspect in an unambiguous way.

Agreement of Judgments The analyses of correlations between latent variables for students and parents show that the degree of agreement is quite different over the four aspects. The correlations for the global concept of empathy (.32 vs. .43) demonstrate a relatively high degree of agreement. These results can be compared to an average of single item correlations of .27, between self-reports and judgments of personality provided by close acquaintances, reported by EJPA 17 (1), © 2001 Hogrefe & Huber Publishers

46

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

Funder and Colvin (1988). They suggested that “psychologists have persistently underestimated the importance of correlations in this range” (p. 156). Agreement is, however, not the same as accuracy, “It is reassuring, therefore, that a survey of the empirical data in this area also supports the existence of a connection.” (Funder & Colvin, 1997, p. 639–640). Because accuracy is fundamental to construct validity (Cronbach & Meehl, 1955) and can never be assessed directly, it can only be established through multiple criteria. In previous research, many of the same circumstances that are likely to enhance accuracy have been shown to improve interjudge agreement. It has been shown that traits which are more visible and higher in terms of social desirability, tend to be judged with better agreement. It has also been shown that knowing a person better improves accuracy (cf. Funder, 1980; Founder Colvin, 1988). Furthermore, thoughts and feelings appear to be more accurately judged by friends than by strangers (Colvin, Vogt, & Ickes, 1997; Stinson & Ickes, 1992). In view of the statements presented above there are several reasons to believe that the agreement between students and parents that emerged from the present study is accurate. The first, and probably most important, reason is the subjects’ specific parent-adolescent relationship (Funder & Colvin, 1997). A second reason concerns trait properties. There is reason to assume that empathy is both a socially desirable and a relatively visible concept. The results from the mean comparisons of the global empathic factor showed significant differences with a higher level for the parents, which is in accordance with previous findings (e. g., Funder, 1980; Funder & Colvin, 1988). Finally, a third reason concerns methodological issues. The correlation estimates are based on reliable variance common to the latent variables (Maruyama, 1997), and when agreement is not found one important reason is unreliable rating scales (Funder & Colvin, 1997). Furthermore, the findings obtained from the replicated investigation of the validity, as accounted for above, confirm the validity of the variables. However, the present study must be sensitive to the presence of extraneous influences such as stereotypic accuracy, elevation and differential elevation (Cronbach, 1955; Cronbach et al., 1972). Considering the quite different correlation estimates obtained over the latent variables this study does not seem to be artifactually affected by shared response dispositions. Regarding stereotypic accuracy, there are reasons to believe that each studentparent pair is similar in personality and in other factors. Thus the correlations may have been inflated by such a phenomenon that they share various perceptual and cognitive biases. There is also reason to believe that parents share some biases while adolescents share others. This kind of artifact implies that the correlations will be atEJPA 17 (1), © 2001 Hogrefe & Huber Publishers

tenuated. Artifacts may have a greater influence on the estimates if the content of the items is hard to interpret and/or difficult to judge. These kinds of artifacts should reasonably also imply some amount of random noise, which is sorted out by using confirmatory factor analysis (Gustafsson & Stahl, 1997). Effects of responses that really manifest shared biases between judge and target are, however, more problematic, although this kind of phenomenon, “. . . is difficult to support, however, because the biases said to operate in trait attribution generally affect the perceptions of the observer, but not the actor.” (Funder, 1980, p. 488). It is also possible that both interrater- and intrarater-similarities are real and not an effect of biases at all (see Hofstee, 1994). The students and their parents were assured absolute anonymity, in order to minimize any tendency they might have to describe themselves and their adolescents in a favorable light. Therefore the study is burdened with a relatively great number of missing scores from parents. For that reason the validity in the study was tested by performing missing-data modeling (Arbuckle, 1996; Muthén, Kaplan, & Hollis, 1987). The substantial correlation between the students’ data and their parents’ data can be considered as a good basis for such an approach, which implies that the information in the available data is used to predict the missing values. This circumstance strengthens the conclusion from the validity tests that the missing scores have no impact on the main results that emerged from the study. The understanding of the definition and measurement of the concept of empathy has a particular relevance to many areas of psychological research, such as judgmental situations and selection procedures. Judgments of personality always involve such issues as construct validity and accuracy, which never can be assessed directly. The only way to solve this is to accumulate evidence from data of different sorts, gathered and analyzed through diverse and independent methods. Further research may involve not only replications of the present results by the same methods with the same measures, but also other methods (e. g., judgments by strangers and interviews) and additional measures theorized to relate or otherwise to the employed measures used when comparing components.

References Arbuckle, J.L. (1996). Full information estimation in the presence of incomplete data. In G.A. Marcoulides & R.E. Schumacker (Eds.), Advanced structural equation modeling. Issues and techniques. Mahwah, NJ: Erlbaum.

C. Cliffordson: Parents’ Judgments and Students’ Self-Judgments of Empathy

Arbuckle, J.L. (1997). Amos Users’ Guide. Version 3.6. Chicago, IL: SmallWaters Corporation. Bollen, K., & Long, J.S. (1993). Testing structural equation models. Newbury Park, CA: Sage. Cliffordson, C. (2000). The structure of empathy: An analysis of the Interpersonal Reactivity Index (IRI). Manuscript submitted for publication. Colvin, C.R., Vogt, D., & Ickes, W. (1997). Why do friends understand each other better than strangers do? In W. Ickes (Ed.), Empathic accuracy. London: Guilford. Cronbach, L.J. (1955). Processes affecting scores on “understanding of others” and “assumed similarity”. Psychological Bulletin, 52, 177–193. Cronbach, L.J., Gleser, G.C., Nanda, H., & Rajaratnam, N. (1972). The dependability of behavioral measurements. New York: Wiley. Cronbach, L.J., & Meehl, P.E. (1955). Construct validity in psychological tests. Psychological Bulletin, 52, 281–302. Davis, M.H. (1980). A multidimensional approach to individual differences in empathy. JSAS Catalog of Selected Documents in Psychology, 10, 85. Davis, M.H. (1983). Measuring individual differences in empathy: Evidence for a multidimensional approach. Journal of Personality and Social Psychology, 44, 113–126. Davis, M.H. (1996). Empathy – A social psychological approach. Boulder, CO: Westview Press. Funder, D.C. (1980). On seeing ourselves as other see us: Selfother agreement and discrepancy in personality ratings. Journal of Personality, 48, 473–493. Funder, D.C. (1987). Errors and mistakes: Evaluating the accuracy of social judgment. Psychological Bulletin, 101, 75–90. Funder, D.C., & Colvin, C.R. (1988). Friends and strangers: Acquaintanceship, agreement, and the accuracy of personality judgment. Journal of Personality and Social Psychology, 55, 149–158. Funder, D.C., & Colvin, C.R. (1997). Congruence of others’ and self-judgments of personality. In R. Hogan, J. Johnson & S. Briggs (Ed.), Handbook of personality psychology (pp. 617–647). London: Academic Press. Gustafsson, J.-E., & Stahl, P.A. (1997). STREAMS User’s guide, Version 1.7. for Windows. Mölndal, Sweden: MultivariateWare. Gustafsson, J.-E., & Stahl, P.A. (1998). STREAMS User’s guide, Version 2.0. for Windows. Mölndal, Sweden: MultivariateWare. Hofstee, W.K.B. (1994). Who should own the definition of personality? European Journal of Personality, 8, 149–162. Hogan, R. (1969). Development of an empathy scale. Journal of Consulting and Clinical Psychology, 33, 307–316. Johnson, J.A., Cheek, J.M., & Smither, R. (1983). The structure

47

of empathy Journal of Personality and Social Psychology, 45, 1299–1312. Jöreskog, K.G., & Sörbom, D. (1996). LISREL 8 user’s reference guide (2nd ed.). Chicago: Scientific Software. Kulich, K.M., & Bengtsson, H. (1997). Swedish dental students’ and dentists’ self-perceived interpersonal skills measured by self-report inventories. Manuscript in preparation. Kunda, Z. (1987). Motivated inference: Self-serving generation and evaluation of causal theories. Journal of Personality and Social Psychology, 53, 636–647. Litvack-Miller, W., McDougall, D., & Romney, D.M. (1997). The structure of empathy during middle childhood and its relationship to prosocial behavior. Genetic, Social, and General Psychology Monographs, 123, 303–324. Maruyama, G.M. (1997). Basics of structural equation modeling. London: Sage. Mehrabian, A., & Epstein, N.A. (1972). A measure of emotional empathy. Journal of Personality, 40, 525–543. Muthén, B. O., Kaplan, D., & Hollis, M. (1987). On structural equation modeling with data that are not missing completely at random. Psychometrika, 52, 431–462. Pedhazur, E.J., & Pedhazur Schmelkin, L. (1991). Measurement, design, and analysis: An integrated approach. Hillsdale, NJ: Erlbaum. Riggio, E., Tucker, J., & Coffaro, D. (1989). Social skills and empathy. Personality and Individual Differences, 10, 93–99. Snow, R.E., Corno, L., & Jackson III, D. (1996). Individual differences in affective and conative functions. In D.C. Berliner & R.C. Calfee (Eds.), Handbook of educational psychology (pp. 243–310). New York: Simon & Schuster Macmillan. Stinson, L., & Ickes, W. (1992). Empathic accuracy in the interactions of male friends versus male strangers. Journal of Personality and Social Psychology, 62, 787–797. Wise, P.S., & Cramer, S.H. (1988). Correlates of empathy and cognitive style in early adolescence. Psychological Reports, 63, 179–192.

Christina Cliffordson Trollhättan/Uddevalla University P.O. Box 1236 S-46228 Vänersborg Sweden Tel. +46 521 264228 Fax +46 521 264298 E-mail [email protected]

EJPA 17 (1), © 2001 Hogrefe & Huber Publishers