FAIRNESS EVALUATION

In this section, I analyzed the second dependent variable: the participants’ fairness evaluations. After submitting their first contribution, participants were told that their partner contributed only half of what the participant gave to the group account. Then, they were required to evaluate the fairness of the situation (scaled as 1= very unfair, 2= unfair, 3= somewhat unfair, 4= indifferent, 5= somewhat fair, 6= fair, and 7= very fair).

Data

The data are slightly non-normal in accordance with the Kolmogorov-Smirnov and Shapiro-Wilk tests. However, the skewness is = 1.82 and kurtosis is = -1.63. Therefore, I conclude that my data display no skewness or kurtosis issues. The data satisfied the assumption of homoscedasticity in both the Breusch-Pagan [LM= .809, p = .369] and Konker [LM= 1.221, p = .269] tests. Also, the data satisfy the assumption of multicollinearity. Consequently, I conducted parametric tests to analyze my data.

Findings

Framing Effect. An ANCOVA test was conducted to determine if there was a

statistically significant difference between differently named tasks on fairness evaluations when statistically controlling for the participants’ initial contribution (i.e. first contribution). The test reveals the effect was in significant [F (2,142) = .231, p = .794]. This means that holding the initial contribution constant, task name did not predict fairness

evaluation. The mean fairness value was 3.39 for the decision group (SD = 1.38), 3.39 for the community group (SD = 1.45), and 3.22 for the Wall Street group (SD = 1.57).

The result indicates that hypothesis 2.a: “Injustice will be stronger in the community

group than the decision group” and hypothesis 2.b: “Injustice will be stronger in the

decision group than the Wall Street group” are not supported by the data.

SVO Effect. I also checked for an effect of SVO, but I did not detect any significant

effect of SVO on fairness evaluation when statistically controlling for the initial contribution. An ANCOVA test shows the effect was insignificant [F (1,119) = 1.306, p = .255] which means that holding the initial contribution constant, SVO did not predict fairness evaluation. The mean fairness value was 3.09 (SD= 1.28) for prosocial participants and 3.47 (SD = 1.58) for individualistic participants. Although prosocial participants showed slightly more anger towards unfair partner than individualistic participants, the difference was not statistically significant.

SVO-Framing Interaction Effect. I also did not find any interaction effect between

task name and SVO when statistically controlling for initial contribution. The results from a two-factorial ANCOVA test are summarized in Table 6.3 below.

Table 6.3: Two-factorial ANCOVA Results for Fairness Evaluation

F P-Value Control Variable: - First Contribution _{.666 .416} Independent Variables: - Task Name _{.014 .986} - SVO _{1.317 .254} - Task Name*SVO _{.706 .496} * _{= p ≤ .10,} ** _{= p ≤ .05,} *** _{= p ≤ .01,} **** _{= p ≤ .001.}

First Contribution Effect. Since I could not find any significant effect for framing and SVO on fairness evaluation, I checked whether the participant’s own first contribution influenced their fairness evaluation. In other words, I checked whether or not participants who gave more money to the group account in the first round evaluated the situation as more unfair than participants who gave less money. A linear regression shows this effect was insignificant [t = -1.546, p = .124] which means that participants’ initial contribution did not predict fairness evaluations.

First Contribution-Framing Interaction Effect. Framing and participants’ first

contribution were insignificant in predicting fairness evaluation as single factors. I also checked whether or not how much participants gave to the group account in the first round and which task they were assigned to had any interaction effect on fairness evaluation. To see the interaction effect, I conducted a hierarchical linear regression and incorporated multiple predictors. The models are summarized in Table 6.4 below.

Table 6.4: Hierarchical Multiple Regression Models for Fairness Evaluation

Model 1 B (SE) Model 2 B (SE) Model 3 B (SE) Intercept 3.220 (.208) **** 3.559 (.302) **** 4.354 (.411) **** Independent Variables: - Decision1 _{.168 (.296) .178 (.294) -1.219 (.585) **} - Community1 _{.171 (.300) .169 (.299) -.890 (.620)} - First Contribution _{-.059 (.038) -.199 (.063) ***} - Decision*FirstCont _{.241 (.088) ***} - _{Community*FirstCont .186 (.096) *} Omnibus F Tests .218 .942 2.213* * _{= p ≤ .10,} ** _{= p ≤ .05,} *** _{= p ≤ .01,} **** _{= p ≤ .001.} 1 _{The Wall Street task is the reference category.}

As can be seen from Model 1 and Model 2, when task name and first contribution were in the model as predictors, their main effects were insignificant. However, task name and initial contribution interacted significantly, as seen in Model 3. This means that initial contribution was moderated by the task name variable in predicting fairness evaluation. Because these factors interacted, simply examining their main effects can lead to incorrect conclusions (Baron and Kenny 1986; Hayes 2013). Although I expected that people who contributed more money to the group account in the first round would experience stronger injustice, statistical tests show that this effect was insignificant. However, the interaction effect leads me to check whether this relationship varies across differently named tasks and is more complicated than the interpretation of the main effects would suggest.

Partitioned Data Analysis. After finding a significant interaction effect between

framing and first contribution (See Table 6.4), I partitioned my data by task name, and looked closer at the data. Compared to the other tasks, the Wall Street group was the only task where people who gave more money in the first round evaluated the situation as more unfair than people who gave less money [t= -3.156, p = .003]. Low-givers and high-givers evaluated the situation very similarly in the community and decision groups. The regression test results were statistically insignificant for the community group [t= -.171, p =.865] and the decision group [t= .716, p = .477].

Overall, the partitioned data and interaction analyses show that task name and first contribution were important factors in predicting fairness evaluation. Framing did predict fairness evaluation only if initial contribution was taken into account in a model. However, these results do not support hypothesis 2.a: “Injustice will be stronger in the community

decision group than the Wall Street group”. Actually, the results indicate that participants in the Wall Street task were more likely to be more sensitive to unfair partner than participants in the community and decision tasks when their own initial contribution was taken into account. In other words, the results determine that participants who gave more money to the group account and were assigned to the Wall Street task evaluated the situation as more unfair than others.

Additional ANOVA Tests. Additionally, I transformed the first contribution variable

into a categorical variable which is coded as “low-giver” if the contributed amount is less than $5.00, “moderate-giver” if the contributed amount is equal to $5.00, or “high-giver” if the contributed amount is more than $5.00. Then, I ran a two-factorial ANOVA test to analyze the interaction effect of the task name and first contribution variables. The results are summarized in Table 6.5 below.

Table 6.5: Two-factorial ANOVA Results for Fairness Evaluation

F P-Value

Independent Variables:

- Task Name _{.053 .948}

- (Categorical) First Contribution _{.912 .404} - Task Name*(Categorical) FirstCont _{1.498 .206}

* _{= p ≤ .10,} ** _{= p ≤ .05,} *** _{= p ≤ .01,} **** _{= p ≤ .001.}

I obtained the estimated marginal means from the previous ANOVA test and generated Figure 6.1 to show the marginal means of fairness evaluations by study name and the participants’ contribution level in the first round.

Figure 6.1 shows that fairness evaluations vary significantly by giving-level only in the Wall Street group. The difference between low-givers and high-givers within the Wall Street group is 1.32, but only .25 in the community group and -.44 in the decision

group. This indicates that the Wall Street group evaluated fairness according to how much they gave to the group account.

Figure 6.1: Estimated Marginal Means of Fairness Evaluation