1. Background Information
Author: Batson, Daniel C., Shannon Early and Giovanni Salvarani
Title: Perspective Taking: Imagining How Another Feels Versus Imagining How You Would Feel
Source: Personality and Social Psychology Bulletin (PSPB), Vol. 23, No.7, pp 741-758.
Year: 1997

2. Abstract
Literature review showed that the current empathy literature took it for granted that the psychological process and empathic emotion triggered by adopting two different perspectives (imagine-other vs. imagine-self) are the same. The authors wanted to demonstrate in their article (the first research purpose) that it was not the case and hypothesized that adopting an imagine-other perspective can trigger empathy to the person in need whereas adopting an imagine-self perspective can trigger both empathy to the person in need and distress. The second research purpose is the demonstration of and solution to a current methodological problem in empathy literature. The current procedure of measuring distress does not distinguish between self-oriented distress and other-oriented distress (a feeling of distress for others). The authors argued that the latter form of distress should belong to empathy rather than the direct personal distress and such confusion is a threat to content validity of distress measures. The authors conducted two separate two-way mixed-factor ANOVAs (3*2) to address the above two questions. 60 undergraduates participated the experiment although 4 were excluded from further statistical analysis because of their suspicion of authenticity of experimental manipulation. They were assigned to three treatment conditions and were asked to listen to a bogus pilot radio interview with a young woman in serious need and then were required to indicate their emotional response. While listening to the radio, they were asked to take different perspectives dependent on the treatment (imagine-other, imagine-self and objective). The results showed support to both hypotheses. Specifically, imagining how the person in need feels in the need situation produced empathy whereas imagining how a subject himself or herself would feel if he or she were in the need situation produced both empathy and personal stress. What is more, the existent measurement of distress did confuse two types of distress - direct personal distress versus distress for others.

3. Null Hypotheses, alpha (or p) level, and sample size per group
First all of all, I want to claim that I do not think the authors used a correct design (to be discussed later in this question) to answer either of their research questions. Second, the authors did not provide specific information regarding the dependent variable in either of the research questions (to be discussed in question 4). Virtually the two sets of statistical analysis and designs face similar problems. The research hypotheses listed as follows were based on the assumption that the authors were correct and that the authors used empathy index as dependent variable in the first question and direct distress index as dependent variable in the second research question.
The first set of null hypotheses (related to the first research question: adopting an imagine-other perspective can trigger empathy to the person in need whereas adopting an imagine-self perspective can trigger both empathy to the person in need and distress):
H01: The specific perspective the subjects take (imagine-self perspective, imagine-other perspective or objective perspective) when they listen to the taped radio interview with a person in need will not influence the degree to which the subjects experience empathic emotion (their scores on empathy index).
H01: j = 0 for all j (in this case j= 1, 2, 3)

H02: The type of vicarious emotion evoked (empathy or distress) will not influence subject's scores on empathy index (the degree to which the subjects experience empathic emotion).
H02: k= 0 for all k (in this case k= 1, 2)

H03: There will be no influence on the degree to which the subjects experience empathic emotion associated with the joint effects of the specific perspective the subjects adopt while listening to the taped radio interview and the type of vicarious emotion evoked. In other words, the difference between subjects' experienced empathic emotions due to the different perspective they adopt should not be dependent on the specific types of vicarious emotions evoked.
H03: ( ) jk = 0 for all j and k (in this case j= 1, 2, 3 and k= 1, 2)
Actually I believe there are problems associated with the second and third null hypothesis. The authors treated perspective taking as a between-subject variable and the type of emotion evoked as a within-subject variable. It is known from prior discussion that the first research purpose is to study the relationship between perspective taking and the types and degrees of vicarious emotion evoked. Therefore, the type of vicarious emotion evoked should be the dependent variable. However, the authors treated it as an independent variable. Besides, the second hypothesis is of no theoretical interest at all. Without carrying out statistical analysis, we will definitely expect a significant main effect of the type of vicarious emotions evoked on the degree to which the subjects experience empathic emotions. Another way to explain this main effect is that when the emotions evoked after listening to the tape is empathy, the subjects will score high on empathy index, which is of common sense and not at all worth studying. Since it is not correct to specify type of vicarious emotions evoked as independent variable, it makes no sense to interpret the interaction effect between the two independent variables.
The second set of null hypotheses (related to the second research question to further decompose distress: the type of distress evoked when subjects adopt the imagine-other perspective should be other-oriented and belong to empathic distress, whereas the type of distress evoked when subjects adopt the imagine-self perspective should be a mix of other-oriented distress and direct distress):
H04: The specific perspective the subjects take (imagine-self perspective, imagine-other perspective or objective perspective) when they listen to the taped radio interview with a person in need will not influence the degree to which the subjects experience direct personal distress (their scores on direct distress index).
H04: j = 0 for all j (in this case j= 1, 2, 3)

H05: The type of distress evoked (direct or for Katie-the person in need) will not influence subject's scores on the direct distress index (the degree to which the subjects experience direct personal distress).
H05: k= 0 for all k (in this case k= 1, 2)

H06: There will be no influence on the degree to which the subjects experience direct personal distress associated with the joint effects of the specific perspective the subjects adopt while listening to the taped radio interview and the type of distress evoked. In other words, the difference between subjects' experienced direct personal distress due to different perspectives they adopt should not be dependent on the specific types of distress evoked.
H06: ( ) jk = 0 for all j and k (in this case j= 1, 2, 3 and k= 1, 2)
This set of hypothesis faces the same problem as the first set of hypothesis. Namely, I think the fifth and sixth hypotheses are problematic. The authors treated the type of distress evoked as within-subject variable. It can be drawn that the second research purpose is to study the relationship between perspective taking and the types and degrees of distress evoked. Therefore, the type of distress (direct vs. for Katie) evoked should be the dependent variable. However, the authors treated it as independent variable. Besides, the fifth hypothesis is of no theoretical interest at all. It is basically saying that when the distress evoked after listening to the tape is self-oriented (person direct stress) the subjects will score high on direct distress index, which is of common sense and not at all worth studying. Since it is not correct to specify type of distress evoked as independent variable, it makes no sense to interpret the interaction effect between the two independent variables.
Pre-set alpha/ p-value: The alpha was not stated in either the research design or in the analysis section of the article. However, the authors did report p-level associated with each effect. The p values of significant effects range from 0.0001 to 0.05.
Sample size per group: The participants of this experiment are 60. The authors stated that in each cell (3 levels of perspective-taking), there are 10 males and 10 females. The authors mentioned that 4 data (all male) were excluded from further analysis because their doubt about the authenticity of the taped interview. (2 from the imagine-other condition, 1 from each of the other conditions) Therefore, the sample size is 56 (N=56).
It is also worth mentioning that the authors originally used a randomized block procedure to assign the subjects. Although it is the only place in the article mentioning this assignment procedure, it can be inferred that the authors might originally want to utilize a randomized block design and control the blocking variable of sex (actually it makes sense to expect an sex effect on empathy emotion theoretically). However, since later statistical analysis failed to show support of this hypothesis (pp754), they had to give up the randomized block design. (All the above analyses were not stated in the article besides the mentioning of "randomized block procedure" to assign subjects and the insignificant sex effect). The authors should make clear this process since it may cause confusion to the readers when they try to figure out the underlying theoretical reasons to utilize this "randomized block procedure" to assign subjects.
The final sample size per cell is shown as follows (perspective-taking as between-subject variable, types of vicarious emotions evoked as within-subject variable):
Imagine-other Imagine-self Objective
Empathy 18 19 19
Distress

However, in later analysis, the authors did not state how they solved the problem of unequality of cell sample size.

4. Independent variable and dependent variables
Although I think the authors are erroneous in terms of their basic design form, for the sake of analysis, let us first assume the authors were right. In the first mixed- factor ANOVA, the independent variables are perspective taking and types of vicarious emotions evoked. The dependent variable is the empathy index.
Perspective-taking was treated as a fixed and between-subject independent variable. The first level is objective perspective. In this level, subjects were required to view the person in need from an objective and detached perspective and were discouraged from imagining from the target person's perspective. The second level is imagine-other perspective. In this level, the subjects were required to imagine how the target person feels in the need situation. The third level is imagine-self perspective. In this level, the subjects were required to imagine how they themselves would feel if they were in the target person's situation.
According to the authors, the second independent variable is the type of vicarious emotion evoked, which was treated by the authors as a fixed and within-subject independent variable. The first level of this variable is empathy, which reflects an other-oriented emotional response congruent with the perceived plight of the person in need. The second level of this variable is distress, which is defined as a self-oriented aversive emotional response and taps more direct feelings of discomfort evoked by witnessing the plight of the other. Just as I have discussed before, I think that the authors were erroneous to treat the variable as an independent variable.
As I have mentioned before, the authors did not specify clearly which was considered as dependent variable. I think there are three possibilities of dependent variable. The first possibility is overall emotional reaction index (my label, the authors did not state it), which is the average of subject's scores on the total 26-item emotional reactions scale. The second possibility is empathy index, which is the average of the 6 items concerning empathy among the overall 26 items. The third possibility is the average of the 8 items concerning distress among the overall 26 items. The authors did not tell the readers which the above three alternatives is the one they used as dependent variable.
The second so-called "mixed"-factor ANOVA faces the same problems. Consistent with the analysis of first ANOVA, let us first assume the authors were right.
The first independent variable, perspective-taking, is the same as the first ANOVA. The second independent variable, type of distress, is a fixed and within- subject variable. It has two levels. The first level of this independent variable is direct distress, which refers to the kind of distress that is directly experienced by people after they hear the plight of the person in need. The second level is empathic distress, which refers to other-oriented emotional response congruent with the perceived welfare of another.
The dependent variable again faces the same problem as the first ANOVA. I think there are two possibilities of dependent variable. The subjects were asked to indicate the degree they personally directly experienced each of these types of response for four of the distress adjectives rated previously (distressed, upset, troubled and grieved) and the degree to which they feel each of the same four distress adjectives for the person in need. The scores were averaged to form a direct stress index and empathic stress index. It is not clear that which of the two indexes the authors used as dependent variables.
Again, I think the authors were erroneous to treat types of distress evoked as an independent variable. According to author's research hypotheses, the type of distress should be reflections of the manipulations of perspective taking.

5. Instrument, comment on its reliability and validity.
The first instrument/questionnaire the authors used was to measure the emotional reaction to Katie's need. After listening to the tape, participants completed the reaction questionnaire. Part 1 of the questionnaire listed 26 adjectives describing different emotional states. For each item, the subjects indicated on a 7-point likert scale to indicate how much they had experienced that emotion while listening to the tape. The list included the six items concerning empathy. It also included eight items concerning distress. However, the authors did not mention in their article the remaining 12 items. Part 2 of the reaction questionnaire was designed to assess the nature of distress reported in Part 1. Participants were asked to indicate on a 9 point likert scale the degree to which they experienced each of these types of response for four of the distress items rated in Part 1: distress, upset, troubled and grieved. They also were required to indicate the degree to which they feel the same types of distress for Katie (the person in need).
In part 1, the six items related to empathy were averaged to form an empathy index and the eight items related to distress were averaged to form a distress index. In part 2 of the research questionnaire, the four items related to distress for self (direct distress) were averaged to form a direct distress index and the four items related to distress for Katie were averaged to form a distress-for-Katie index.
I would take issue with the authors here because I think items of part 2 are subject to demand characteristics. It would be very easy for the subjects to guess the purpose of the researchers. For example, when the subjects have just read in the instructions that they are required to think from the perspective of Katie (how she will feel in such situation) and later the subjects were required to indicate how much distress for the self vs. for Katie, it is very easy for the subjects to identify the research hypotheses of the researchers and therefore they tend to give a high score on the distress-for-Katie items. The direction of the effect of demand characteristics on the dependent variable is the same as the research hypothesis, which causes potential confounding effects to the conclusion of the researchers.
The authors provided reliability for the four indexes. The Cronbach's alphas for the four indexes-empathy, distress, direct distress and distress- for -Katie are: 0.85, 0.93, 0.97 and 0.95, respectively, which are all beyond traditional rules-of-thumb 0.7.
However, the authors did not provide any evidence regarding validity of the four indexes. Information such as content validity, criterion-related validity, construct validity (convergent and discriminant) validity were not provided.
The authors also used what they called "ancillary measures" in order to investigate the effect of manipulations of perspective taking. They also required the subjects to indicate how interesting and worthwhile the radio program was in order to test the believability of the radio interviews. Since there is only one item associated with each question, no information regarding reliability and validity were provided, which are acceptable.

6. Experimental Procedure
Sixty psychology undergraduates participated the experiment. They were recruited and were told that they were involved in part of a pilot test of new programming ideas for the local university public radio station. According to the authors, they used a "randomized block procedure" to assign 10 women and 10 men to each of three perspective-taking conditions. As we know, the randomized block procedure is utilized only when the researchers suspect that there might be a nuisance variable (blocking variable) that they want to control for because of the potential confounding effect of this nuisance variable on dependent variable. However, the authors did not mention the potential effect of gender on the dependent variable (empathy vs. personal distress) at all, let alone providing any theoretical support to this potential effect.
4 additional data (2 from the imagine-other condition; 1 from each of the other conditions) were excluded because the subjects doubt about the credibility of the manipulation.
In the instructions contained in a folder that the subjects received upon arrival (the experimenter had no idea what specific contents were included), the participants were told to take a particular perspective, dependent on which condition they were assigned to - imagine other, imagine self and objective. In the imagine-other condition, the subjects were told that they should imagine how the person being interviewed feels about what has happened and how it has affected his or her life. In the imagine-self condition, the subjects were told to imagine how they themselves would feel if they were experiencing what has happened to the person being interviewed and how this experience would affect their life. In the objective condition, the subjects were told to be as objective as possible about what has happened to the person interviewed and how it has affected his or her life.
After the instructions, participants were left alone and listened to the radio interview with Katie Bank, whose parents were both killed recently in a car accident and who were trying her best to finish college and take care of her surviving brother and sister.
After listening the tape, the subjects were asked to complete a questionnaire, which were consistent of two parts. The 26 items contained in the first part of questionnaire were designed to measure the emotional reactions of the subjects. The 8 items contained in the second part of the questionnaire were designed to assess the nature of any distress reported in part 1. The subjects also completed some ancillary measure designed to assess the effect of manipulation and authenticity of the radio interview. In the end, the experimenter returned and conducted a full debriefing.

7 Statistical analysis and conclusion
In order to test their first hypothesis that adopting an imagine-other perspective will lead to empathy and adopting an imagine-self perspective will lead to both distress and empathy, the authors conducted a 2 (type of vicarious emotion: empathy, distress) * 3 (perspective-taking condition: objective, imagine other, imagine self) mixed-factor ANOVA with repeated measures on the first factor.
Just I have discussed above, there are two major drawbacks with this data analysis procedure. The first problem is that the authors should not treat the types of vicarious emotion evoked as independent variable. From a theoretical point of view, the types of vicarious emotion evoked should be a result (dependent variable) of the manipulation of another independent variable-perspective taking. From an operationalization (manipulation) point of view, the manipulation of one factor should not tap into the construct domain of the other factor. As we know, the types of vicarious emotion evoked are reflections of the manipulation of perspective taking. Therefore, we can expect that the types of vicarious emotions should be highly correlated to perspective taking. In other words, the manipulation of perspective taking will not only tap into the construct the manipulation ought to influence, it also taps into types of vicarious emotions evoked, the other factor. The contamination of manipulation may call the validity of researchers' conclusion into question. The second problem is that the authors did not provide any information regarding which is the dependent variable-empathy index, distress index or overall emotion index.
For the sake of analysis, let us assume that the authors use empathy index as dependent variable . According to the authors, there was a main effect for types of vicarious emotions on the dependent variable, the empathy index, F (1,57) = 22.21, p <.005. This main effect is so obvious that we do not need to carry out any statistical analysis to test it and it is of no theoretical interests at all. The main effect for perspective taking on empathy index was significant, F(2, 57) = 8.10, p<.001. This main effect is of theoretical interest and it tells us that the perspective the subjects take determines the degree to which they feel empathy. However, in order to test which perspective can evoke more empathy, post-hoc analysis should be conducted. The interaction effect of types of vicarious emotion and perspective taking is significant, F (2,57)= 3.24, p<. 05 However, just I have discussed in question 3, since the types of vicarious emotion should not be treated as an independent variable, it makes no sense to interpret this interaction effect.
The authors also conducted a few analyses to prove their research hypothesis. The authors conducted a planed comparison and the results indicated participants in the imagine conditions reported feeling more empathy than did participants in the objective condition, t (57) = 2.76, p <0.01. In order to test the hypothesis that imagine-self perspective would evoke greater distress than the objective and the imagine-other perspectives, the authors conducted planned contrasts and the results indicated that participants in the imagine-self condition reported feeling more distress than did participants in the objective and imagine-other conditions, t (57) = 4.22, p<.0005, with no reliable difference between the objective and imagine-other conditions, t<1.47. What is more, in the imagine-other condition, empathy was significantly higher (M=4.96) than distress (M= 3.99), t (19) = 3.97, p<0.001. In the imagine-self condition, both empathy (M=5.36) and distress (M=2.50) were high, with no reliable difference, t (19) <0.64. Actually I think these procedures are enough to prove their hypotheses. In other words, the authors do not need to go to the mixed-factor ANOVA analysis, which turns out to be erroneous any way. With this set of analysis, we are able to know that empathy was associated with taking an imagine-other perspective, whereas adopting an imagine-self will lead to both distress and empathy.
Although the authors did report p-level associated with the significant results, they failed to report the p-levels associated with the insignificant results.

In order to test their second hypothesis that the nature of distress experienced when adopting an imagine-other perspective will be other-oriented and the nature of distress experienced when adopting an imagine-self perspective will be a mix of self-oriented and other-oriented distress, the authors conducted a 2 (type of distress: direct vs. Katie) * 3 (perspective-taking condition: objective, imagine other, imagine self) mixed-factor ANOVA with repeated measures on the first factor.
The problems with the utilization of mixed-factor analysis here are almost the same as those in the first analysis so they are omitted here to avoid redundancy. For the sake of analysis, let us assume that the authors use direct distress index as dependent variable . According to the authors, there was a main effect for types of distress on the dependent variable, the direct distress index, F (1,57) = 98.73, p <.0001. This main effect is again so obvious that we do not need to carry out any statistical analysis to test it and it is of no theoretical interests at all. I would say that it would be extremely bizarre if the main effect of types of distress (direct distress vs. distress-for-Katie) on direct distress score did not turn out to be significant. The main effect for perspective taking on direct distress index was significant, F (2, 57) = 9.98, p<.0005. However, just as I have discussed above, the results are potentially confounded with demand characteristics. The interaction effect of types of distress and perspective taking is significant, F (2,57) = 3.76, p< .03. Since it is erroneous for the authors to treat type of distress evoked as independent variable, to interpret this interaction effect is not appropriate and makes no sense.
The authors also conducted a few analyses to prove their research hypothesis. The authors conducted a planed comparison and the results indicated participants in the imagine conditions reported feeling more distress for Katie than did participants in the objective condition, t (57) = 3.16, p <0.01. In order to test the hypothesis that imagine-self perspective would evoke greater direct distress than the objective and the imagine-other perspectives, the authors conducted planned contrasts and the results indicated that participants in the imagine-self condition reported feeling more direct distress than did participants in the objective and imagine-other conditions, t (57) = 4.50, p<0.005, with no reliable difference between the objective and imagine-other conditions, t<1.0. What is more, participants in both the imagine-other and the imagine-self conditions reported more distress for Katie (Ms = 6.96 and 7.44, respectively) than direct distress (Ms= 3.45 and 5.69, respectively), ts (19) = 7.55 and 3.00, respectively, ps >.0005 and .01, respectively and this difference was significantly greater in the former condition than the latter, F (1,38) = 5.76, p < 0.03. It seems that the set of analysis here is enough to prove the research hypotheses. In other words, the authors do not need to go to the mixed-factor ANOVA analysis, which turns out to be erroneous any way. With this set of analysis, we are able to know that empathy was associated with taking an imagine-other perspective, whereas adopting an imagine-self will lead to both distress and empathy.
Although the authors did report p-level associated with the significant results, they failed to report the p-levels associated with the insignificant results.

8. If you were the researcher, how would you improve the study? Be specific.
I would consider the following modifications:
(1) In terms of the basic form of design, I would certainly give up mixed factor ANOVA design. As discussed before, this design form is erroneous in that the authors treated the dependent variable as a within-subject variable. I think one-way ANOVA and independent t-tests will be sufficient to solve both of the research questions, as what the authors did also in their analysis.
(2) I would definitely show which variable was considered as dependent variable instead of letting the reader guess which is the most appropriate one of all the available alternatives (empathy index, distress index and overall emotion reactions related to the first research question; direct distress index and empathic distress index related to the second research question.)
(3) In terms of sampling procedures, I would specify why I first used a "randomized block procedure". Just I have analyzed above, I would tell the readers that I would expect gender as possible nuisance variable and wish to control it. Although later results did not support this assumption, which may lead me to give up "randomized block design", I should provide such analysis in front of the readers rather than make them confused.
(4) Since 4 data were missing, which resulted in unequal sample size, I would address the issue how I dealt with the problem of unequal size.
(5) I would state the alpha level in the introduction of research design, which will make the associated p-value interpretable.
(6) I would check the basic assumptions of ANOVA. Specifically, I would provide information regarding (a) normality of dependent variable (whether the distributions of emotion indexes are normal); (b) homogeneity of errors (to avoid heteroskedasticity); (c) independence of errors (the errors associated with each observation should not be correlated with each other) Actually, if the authors were correct in using mixed-factor design, they should check the sphericity assumption. However, since a mixed factor design is not appropriate here, it makes no sense to check this assumption.
(7) I would provide information regarding the effect size of the results. Although we can judge from F-test that the hypothesized relationships are significant, we have no idea how strong the hypothesized influences are. It is possible that although the factor can produce a significant influence on the dependent variable, the influence is nominal.