Preliminary results from Empathy intervention Study 1
Alexander Tagesson
2023-09-08
Load packages
Load functions
Load data
Data processing
1. Descriptive data Outgroup
Empathy variables are operationalized as follows:
1 - Empathic reactions:
a) Did you feel empathy with X?
b) Did you feel sympathy with X?
c) Did you feel compassion with X?
Participants answered using a seven point Likert scale going from: 1 - Not at all; 7 - Very much.
2 - Empathic drivers:
a) How badly affected were you by the story about X?
b) How much did you relate to X as you read his story?
c) How motivated are you to help X improve his situation if you are
given the opportunity to do so?
Participants answered using a seven point Likert scale going from: 1 - Not X at all; 7 - Very X.
3 - Empathic beliefs:
a) To what extent do you think that empathy is a limited resource, for
example if you feel a lot of empathy with one person, you will not be
able to feel as much empathy with another person? (This is the
manipulation check)
b) To what extent do you think we can change our ability to be
empathic?
c) After participating in this experiment and learning more about
empathy, how motivated do you feel to try to increase the empathy you
feel in your everyday life?
Participants answered using a seven point Likert scale going from: 1 - Not X at all; 7 - Very X.
1.1 Table over Participants:
| Characteristic | N = 7411 |
|---|---|
| Age | 41 (33, 53) |
| Ethnicity.simplified | |
| Black | 369 (50%) |
| White | 372 (50%) |
| Country.of.residence | |
| United States | 741 (100%) |
| Highest.education.level.completed | |
| Doctorate degree (PhD/other) | 22 (3.0%) |
| Graduate degree (MA/MSc/MPhil/other) | 108 (15%) |
| High school diploma/A-levels | 175 (24%) |
| Secondary education (e.g. GED/GCSE) | 19 (2.6%) |
| Technical/community college | 115 (16%) |
| Undergraduate degree (BA/BSc/other) | 302 (41%) |
| Employment.status | |
| DATA_EXPIRED | 156 (21%) |
| Due to start a new job within the next month | 4 (0.5%) |
| Full-Time | 329 (44%) |
| Not in paid work (e.g. homemaker', 'retired or disabled) | 72 (9.7%) |
| Other | 14 (1.9%) |
| Part-Time | 98 (13%) |
| Unemployed (and job seeking) | 68 (9.2%) |
| 1 Median (IQR); n (%) | |
1.2 Table displaying means for most important variables:
1.3 Scatterplots for Empathic reactions & Donation:
1.3.1 Scatterplots for Empathic reactions & Donation for Control:
1.3.2 Scatterplots for Empathic reactions & Donation for Limited:
1.3.3 Scatterplots for Empathic reactions & Donation for Unlimited:
1.3.4 Scatterplots for Empathic reactions & Donation for Malleable:
1.3.5 Scatterplots for Empathic reactions & Donation for Normative:
1.4. Scatterplots for Empathic drivers & Donation:
1.4.1 Scatterplots for Empathic drivers & Donation for Control:
1.4.2 Scatterplots for Empathic drivers & Donation for Limited:
1.4.3 Scatterplots for Empathic drivers & Donation for Unlimited:
1.4.4 Scatterplots for Empathic reactions & Donation for Malleable:
1.4.5 Scatterplots for Empathic drivers & Donation for Normative:
1.5. Scatterplots for Empathic beliefs & Donation:
1.5.1 Scatterplots for Empathic beliefs & Donation for Control:
1.4.2 Scatterplots for Empathic beliefs & Donation for Limited:
1.5.3 Scatterplots for Empathic beliefs & Donation for Unlimited:
1.5.4 Scatterplots for Empathic beliefs & Donation for Malleable:
1.5.5 Scatterplots for Empathic beliefs & Donation for Normative:
1.6 Correlation matrix for Empathy measures, Donation & Intervention for Outgroup:
1.6.1 Correlation matrix for Control:
1.6.2 Correlation matrix for Limited:
1.6.3 Correlation matrix for Unlimitied:
1.6.4 Correlation matrix for Malleable:
1.6.5 Correlation matrix for Normative:
1.7. Cronbach’s Alpha for Empathic reactions:
##
## Cronbach's alpha for the 'empathic_reactions_df' data-set
##
## Items: 3
## Sample units: 741
## alpha: 0.9321.7.1 Cronbach’s Alpha for Empathic drivers:
##
## Cronbach's alpha for the 'empathic_drivers_df' data-set
##
## Items: 3
## Sample units: 741
## alpha: 0.8121.7.2 Cronbach’s Alpha for Empathic beliefs:
##
## Cronbach's alpha for the 'empathic_beliefs_df' data-set
##
## Items: 3
## Sample units: 741
## alpha: 0.5242. Replication of Hasson et als results
We look at manipulation check and contrasts between conditions for groups Outgroup (i.e. participants that stated that they perceived protagonist as an outgroup member) and Whole group (i.e. all participants).
2.1 Outgroup
Manipulation check:
Background:
“Regarding the manipulation check, an independent-samples t-test revealed a significant effect of condition, t(1, 198) = −4.11, p < 0.001, d = 0.58, such that participants in the unlimited condition believed empathy is unlimited (M = 4.92, SD = 1.89) more than did those in the limited condition (M = 3.9, SD = 1.58).”, Hasson et al(2022).
2.1.1 tibble with mean and sd on manipulation checks between conditions for Outgroup:
## # A tibble: 5 × 3
## condition mean sd
## <chr> <dbl> <dbl>
## 1 Control 4.84 2.25
## 2 Limited 4.34 2.00
## 3 Malleable 4.66 2.21
## 4 Normative 4.65 2.23
## 5 Unlimited 5.33 1.952.1.2 anova to test manipulation check between conditions for Outgroup:
## Df Sum Sq Mean Sq F value Pr(>F)
## condition 4 77 19.209 4.219 0.0022 **
## Residuals 736 3351 4.553
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 12.1.3 Plotting manipulation check for Outgroup:
Contrasting conditions on Empathic reactions
Background:
“To test whether the manipulation influenced empathic reactions in response to each Syrian refugee’s testimony, we ran a repeated-measures ANOVA with testimony order (1–4) as a within-participant variable, and the condition (unlimited vs. limited) as a between-participants variable. We used testimony order as the within-participant variable because all testimonies were presented in a counterbalanced order to rule out the possibility that the content of the testimonies influences the results. We found a significant main effect of condition on empathic reactions, F(1, 198) = 8.93, p = 0.003, d = 0.423. On average, participants in the unlimited condition felt more empathy toward the outgroup members (M = 5.82, SD = 1.43), compared to those in the limited condition (M = 5.18, SD = 1.60). Pairwise comparisons between the effects of limited and unlimited conditions on empathy in each testimony were all significant (Testimony #1: p < 0.001; Testimony #2: p = 0.003; Testimony #3: p = 0.037; Testimony #4: p = 0.013). Moreover, we found a significant Testimony Order × Condition interaction, F(1, 196) = 2.78, p = 0.042, d = 0.41 (Fig. 2). While empathy changed across stories in the limited condition (between testimonies #1 and #4; p = 0.021), empathy remained stable in the unlimited condition, and there were no significant differences across stories (between testimonies #1 and #4; p = 0.917).”, Hasson et al (2022).
Testing H1a - For Empathic reactions we hypothesize the following directional relationships: Malleable == Unlimited > Social Norm > Limited == Control - to test if results from Hasson et als article replicate.
2.1.4 Plotting mean differences on Empathic reactions between Conditions for Outgroup:
2.1.5 Mean difference between Unlimited vs Control:
## [1] -0.04098423Quantile interval:
## 2.5% 97.5%
## -0.2640912 0.1817685Histogram showing Mean difference between Unlimited & Control:
2.1.6 Mean difference for Malleable vs Control:
## [1] -0.1318073Quantile interval:
## 2.5% 97.5%
## -0.3606882 0.0932938Histogram showing Mean difference between Malleable & Control:
2.1.7 Mean difference for Normative vs Control:
## [1] -0.08704801Quantile interval:
## 2.5% 97.5%
## -0.3170253 0.1441882Histogram showing Mean difference between Normative & Control:
2.1.8 Mean difference for Limited vs Control:
## [1] -0.2503176Quantile interval:
## 2.5% 97.5%
## -0.47869445 -0.01976661Histogram showing Mean difference between Limited & Control:
2.1.9 Mean difference for Unlimited vs Limited:
## [1] 0.2093334Quantile interval:
## 2.5% 97.5%
## -0.01802047 0.43670932Histogram showing Mean difference between Unlimited & Limited:
2.1.10 Mean difference for Malleable vs Limited:
## [1] 0.1185103Quantile interval:
## 2.5% 97.5%
## -0.09948274 0.33956760Histogram showing Mean difference between Malleable & Limited:
2.1.11 Mean difference for Normative vs Limited:
## [1] 0.1632696Quantile interval:
## 2.5% 97.5%
## -0.05790945 0.38196276Histogram showing Mean difference between Normative & Limited:
2.1.12 Cohens d for condition Limited & Unlimited:
## [1] 0.2089973Conclusion H1a for Outgroup:
There is a hypothesized mean difference between conditions Control & Limited (.25) and condition Unlimited & Limited (0.21) for Empathic reactions. For the rest of the contrast, HDI does not exclude zero and therefore we conclude no hypothesized mean difference between these conditions for Empathic reactions.
2.2 Whole group
Manipulation check:
2.2.1 tibble with mean and sd for manipulation check for whole group:
## # A tibble: 5 × 3
## condition mean sd
## <chr> <dbl> <dbl>
## 1 Control 4.76 2.21
## 2 Limited 4.43 2.00
## 3 Malleable 4.82 2.11
## 4 Normative 4.72 2.16
## 5 Unlimited 5.32 1.992.2.2 anova to test conditions for manipulation check for whole group:
## Df Sum Sq Mean Sq F value Pr(>F)
## condition 4 72 18.122 4.123 0.00257 **
## Residuals 885 3890 4.396
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 12.2.3 Plotting manipulation check for whole group:
Contrasting conditions on Empathic reactions
Testing H1a - For Empathic reactions we hypothesize the following directional relationships: Malleable == Unlimited > Social Norm > Limited == Control - to test if results from Hasson et als article replicate.
2.2.4 Plotting mean differences on Empathic reactions between Conditions for Whole group:
2.2.5 Mean difference Unlimited vs Control:
## [1] 0.05643339Quantile interval:
## 2.5% 97.5%
## -0.1540099 0.2673820Histogram showing Mean difference between Unlimited & Control:
2.2.6 Mean difference for Malleable vs Control:
## [1] -0.04619456Quantile interval:
## 2.5% 97.5%
## -0.2602826 0.1697739Histogram showing Mean difference between Malleable & Control:
2.2.7 Mean difference for Normative vs Control:
## [1] -0.01747514Quantile interval:
## 2.5% 97.5%
## -0.2171557 0.1876975Histogram showing Mean difference between Normative & Control:
2.2.8 Mean difference for Limited vs Control:
## [1] -0.1407821Quantile interval:
## 2.5% 97.5%
## -0.3452505 0.0655318Histogram showing Mean difference between Limited & Control:
2.2.9 Mean difference for Unlimited vs Limited:
## [1] 0.1972155Quantile interval:
## 2.5% 97.5%
## -0.006665762 0.403572093Histogram showing Mean difference between Unlimited & Limited:
2.2.10 Mean difference for Malleable vs Limited:
## [1] 0.09458758Quantile interval:
## 2.5% 97.5%
## -0.1107544 0.2989768Histogram showing Mean difference between Malleable & Limited:
2.2.11 Mean difference for Normative vs Limited:
## [1] 0.123307Quantile interval:
## 2.5% 97.5%
## -0.08530215 0.33165025Histogram showing Mean difference between Normative & Limited:
2.2.12 Cohens d for Limited & Unlimited:
## [1] 0.1946075Conclusion H1a for Whole group:
There is a hypothesized mean difference between conditions Unlimited & Limited (.2) for Empathic reactions. For the rest of the contrasts, HDI does not exclude zero and therefore we conclude no hypothesized mean difference between these conditions for Empathic reactions.
General conclusion:
We replicate Hasson et als results for both groups, Outgroup and Whole group, that is - there is a hypothesized mean difference between condition Limited & Unlimited for Empathic reactions. Importantly, there is no difference between condition Unlimited & Control, or Unlimited and the other conditions for both groups. This suggests that the difference between Limited & Unlimited is partly driven by a negative impact on Empathic reactions caused by condition Limited. In accordance with this suggestion, we also found a mean difference between Limited & Condition for Outgroup.
3. Modeling Empathic drivers
H1b - For Empathic drivers we hypothesize the following directional relationships: Malleable == Unlimited > Social Norm > Limited == Control.
## Family: gaussian
## Links: mu = identity; sigma = identity
## Formula: empathic_reactions_z ~ 0 + condition
## Data: outgroup (Number of observations: 741)
## Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
## total post-warmup draws = 4000
##
## Population-Level Effects:
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## conditionControl 0.10 0.08 -0.06 0.26 1.00 5827 2859
## conditionLimited -0.15 0.08 -0.31 0.01 1.00 6052 3047
## conditionMalleable -0.03 0.08 -0.19 0.13 1.00 6483 2935
## conditionNormative 0.01 0.08 -0.14 0.18 1.00 5488 3335
## conditionUnlimited 0.06 0.08 -0.10 0.22 1.00 5771 2812
##
## Family Specific Parameters:
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## sigma 1.00 0.03 0.95 1.05 1.00 5951 2809
##
## Draws were sampled using sample(hmc). For each parameter, Bulk_ESS
## and Tail_ESS are effective sample size measures, and Rhat is the potential
## scale reduction factor on split chains (at convergence, Rhat = 1).## Warning: Dropping 'draws_df' class as required metadata was removed.## b_conditionControl b_conditionLimited b_conditionMalleable
## 2.5% -0.1702859 -0.26659882 -0.1605880
## 97.5% 0.1407791 0.05476715 0.1671147
## b_conditionNormative b_conditionUnlimited
## 2.5% -0.05879902 -0.1469587
## 97.5% 0.26525970 0.1753830Mean difference for Unlimited vs Control
## [1] 0.02863183
Quantile interval
## 2.5% 97.5%
## -0.1986312 0.2564561The HDI does not exclude zero, so we conclude no hypothesized difference between condition Control and Unlimitied for Empathic drivers
Mean difference for Malleable vs Control
## [1] 0.01720522
Quantile interval
## 2.5% 97.5%
## -0.2068777 0.2415954The HDI does not exclude zero, so we conclude no hypothesized difference between condition Malleable and Unlimitied for Empathic drivers
Mean difference for Normative vs Control
## [1] 0.1181297
Quantile interval
## 2.5% 97.5%
## -0.1093017 0.3505581The HDI does not exclude zero, so we conclude no hypothesized difference between condition Normative and Unlimitied for Empathic drivers
BRMS model empathic drivers
## Family: gaussian
## Links: mu = identity; sigma = identity
## Formula: donation_z ~ 0 + condition
## Data: outgroup (Number of observations: 741)
## Draws: 4 chains, each with iter = 2000; warmup = 1000; thin = 1;
## total post-warmup draws = 4000
##
## Population-Level Effects:
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## conditionControl -0.01 0.08 -0.17 0.14 1.00 5571 2979
## conditionLimited -0.03 0.08 -0.19 0.13 1.00 5159 2871
## conditionMalleable -0.03 0.08 -0.19 0.14 1.00 5463 2672
## conditionNormative 0.11 0.08 -0.05 0.28 1.00 5427 3302
## conditionUnlimited -0.03 0.08 -0.20 0.13 1.00 5833 3185
##
## Family Specific Parameters:
## Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
## sigma 1.00 0.03 0.95 1.05 1.00 6963 3278
##
## Draws were sampled using sample(hmc). For each parameter, Bulk_ESS
## and Tail_ESS are effective sample size measures, and Rhat is the potential
## scale reduction factor on split chains (at convergence, Rhat = 1).## Warning: Dropping 'draws_df' class as required metadata was removed.## b_conditionControl b_conditionLimited b_conditionMalleable
## 2.5% -0.1688229 -0.1938809 -0.1880629
## 97.5% 0.1423076 0.1282579 0.1353247
## b_conditionNormative b_conditionUnlimited
## 2.5% -0.04986123 -0.1968648
## 97.5% 0.27561948 0.1319767Mean difference for Unlimited vs Control
## [1] -0.01859229
Quantile interval
## 2.5% 97.5%
## -0.2371990 0.2159797The HDI does not exclude zero, so we conclude no hypothesized difference between condition Control and Unlimitied for Donation
Mean difference for Malleable vs Control
## [1] -0.01199492
Quantile interval
## 2.5% 97.5%
## -0.2417528 0.2144014The HDI does not exclude zero, so we conclude no hypothesized difference between condition Malleable and Unlimitied for Donation
Mean difference for Normative vs Control
## [1] 0.1228901
Quantile interval
## 2.5% 97.5%
## -0.1005301 0.3581822