Overview

We are focusing here on caring action/prosocial behavior, rather than on self and other compassion.

Authors

(not in final order)

  • ACU (Australian Catholic University):
    Sneha Ahmed (Masters student leading project), Joseph Ciarrochi, Madeliene Fraser, Baljinder Sahdra
  • University of Nevada, Reno:
    Steven C. Hayes
  • Universität Luzern:
    Andrew Gloster
  • La Trobe University:
    Achini Adikari, Damminda Alahakoon, Isuru Ranapanada
  • Universidad Adolfo Ibáñez:
    Cristóbal Eduardo Hernández Contreras

Key Question

Does caring action lead to well-being?

Review of Literature

The literature suggests that people who engage in caring actions more than average tend to have higher well-being than average. This conclusion is typically seen between individuals.

Can we make conclusions within person based on these between person effrects? Giving has been described as one of the five “ways to well-being.” However, this is a within person claim based on between person effects.

Initial Hypothesis Testing

To address this hypothesis, you can look at the ** pooled correlations** initially.

Findings

  • Within-Person Relationship: There is almost no within-person relationship between caring for others (care_for_other_num) and well-being variables, with all correlations under 0.06.
  • Between-Person Relationship: However, there is a reasonably robust between-person relationship. Those who care more than average have higher well-being on average.

The between and within analyses thus present different conclusions.

Discussion of Null Results

The null result for within-person effects is interesting because, as we will see, the variable is equasyncratic (a new term coined by Steve).

Table: Variability in Caring and Unhappiness

In the table depicting variability for the link between caring and unhappiness, we observe that for some individuals, caring brings more unhappiness, while for others, it brings more happiness. The pooled effect (middle blue bars) shows that the positive and negative effects cancel out, concealing the significant impact caring can have on well-being for many people. The pooled effect makes it seem as if caring has no effect.

Further Analysis

The rest of the analysis attempted to find differences between the two groups—those who experience happiness and those who experience unhappiness when caring. We did not find any significant differences, so more research is needed here

Codebook

Compassionate Action

  • CareForOther = V_19
    CAT since the last scheduled prompt: Did you help, care for, support, or do anything for another person?

  • CareForAltruistic = V_19_1
    CAT: Did this happen without self-interest or profit for yourself and without compensation?

  • WhoCaredFor = V_19a
    CAT: Whom did you help?
    0 - spouse, 1 - children, 2 - other family, 4 - work colleague, 5 - fellow student, 6 - medical staff, 7 - fellow patients, 8 - other

    ** Important. the reason for caring items were reverse coded. I will reverse back so higher numbers mean more of the state or more “bad” feelings **

This was original but then programmer reversed it (andrew email may 29,2024, 1049 am)
Slider from 0 – 100: 0 = not true 100 = true

  • CareVoluntary = V_19_2b
    Why did you help this person? I did it voluntarily.

  • CareObliged = V_19_2c
    I felt obliged to do so. Slider from 0 – 100:0 = not true 100 = true

  • CareGuilt = V_19_2d
    I had a bad conscience or felt guilty.

  • CareAvoidProblem = V_19_2e
    I wanted to avoid problems.

  • CareFeelings = V_19_2f
    How did you feel doing it? Slider from 0 – 100: 0 = very bad 50 = okay 100 = very good

Well-being

  • HowUnhappy = V_10_1
    Since last prompt: 1 to 100 (very much)

  • WithoutEnergy = V_10_2

  • Distracted = V_10_3

  • Stressed = V_10_4

  • AtEase = `V_11_1’ Slider from 0 – 100:0 = not at all 50= somewhat 100 = very much’

  • Optimistic = V_11_2

  • Delighted = V_11_3

  • Satisfied = V_11_4

  • Grateful = V_11_5

Compassion as Perspective Taking (Detection?) we are not looking at this for now

  • SelfCompassion = V_14_1g
    I looked at myself with tolerance, goodwill, and care.

  • OtherCompassion = V_14_1h
    I looked at others with tolerance, goodwill, and care.

Missing data exploration

Descriptives

Skewness values between -0.5 and 0.5 indicate an approximately symmetrical distribution. Values within -1 to -0.5 or 0.5 to 1 suggest a slightly skewed distribution. Values beyond -1 or above 1 are considered highly skewed. Kurtosis quantifies distribution “tailedness.” A kurtosis greater than 3 (excess kurtosis > 0) indicates heavy tails and a more peaked distribution, while less than 3 (excess kurtosis < 0) suggests lighter tails and a flatter distribution

## 
## Female   Male 
##     59     70
## 
##  Inpatient Outpatient 
##         59         70

## # A tibble: 1 × 5
##   Variable  Mean    SD Skewness Kurtosis
##   <chr>    <dbl> <dbl>    <dbl>    <dbl>
## 1 Age       36.2  11.5    0.443   -0.809

## # A tibble: 2 × 5
##   Variable         Mean    SD Skewness Kurtosis
##   <chr>           <dbl> <dbl>    <dbl>    <dbl>
## 1 OtherCompassion  76.6  19.8   -1.33     2.05 
## 2 SelfCompassion   59.6  24.7   -0.453   -0.494
## 
##        cared for another Did not care for another 
##                     1050                     2271
## 
##          Child Fellow Patient Fellow student         Friend  Medical Staff 
##            104            271             15            133             27 
##          Other   Other Family         spouse Work Colleague 
##             90            115            181             95

## # A tibble: 1 × 5
##   Variable          Mean    SD Skewness Kurtosis
##   <chr>            <dbl> <dbl>    <dbl>    <dbl>
## 1 CareForOther_num 0.316 0.465    0.790    -1.38

## # A tibble: 5 × 5
##   Variable           Mean    SD Skewness Kurtosis
##   <chr>             <dbl> <dbl>    <dbl>    <dbl>
## 1 CareAvoidProblemr  20.1  27.1    1.53      1.20
## 2 CareFeelingsr      23.6  16.6    0.850     1.57
## 3 CareGuiltr         14.9  22.3    2.18      4.32
## 4 CareObligedr       40.0  34.0    0.350    -1.37
## 5 CareVoluntaryr     83.0  23.5   -2.02      3.42

## # A tibble: 1 × 5
##   Variable               Mean    SD Skewness Kurtosis
##   <chr>                 <dbl> <dbl>    <dbl>    <dbl>
## 1 CareForAltruistic_num 0.871 0.335    -2.22     2.91

## # A tibble: 9 × 5
##   Variable       Mean    SD Skewness Kurtosis
##   <chr>         <dbl> <dbl>    <dbl>    <dbl>
## 1 AtEase         52.9  25.1   -0.277   -0.631
## 2 Delighted      54.5  25.9   -0.333   -0.669
## 3 Distracted     36.9  26.5    0.348   -0.900
## 4 Grateful       59.8  26.8   -0.505   -0.585
## 5 HowUnhappy     38.7  26.8    0.291   -1.01 
## 6 Optimistic     57.9  24.1   -0.433   -0.391
## 7 Satisfied      58.3  25.0   -0.444   -0.499
## 8 Stressed       40.8  28.1    0.275   -1.04 
## 9 WithoutEnergy  37.6  27.0    0.349   -0.960

Pooled correlations

Nomethetic (group level)

This shows you the simple correlations and significant levels

Within are correlations within person Between are correlations between people

These are pooled effects…they are averages, and hide individual variability.

“CareVoluntary”
[11] “CareObliged” “CareGuilt”
[13] “CareAvoidProblem” “CareFeelings”
[15] “HowUnhappy” “WithoutEnergy”
[17] “Distracted” “Stressed”
[19] “AtEase” “Optimistic”
[21] “Delighted” “Satisfied”
[23] “Grateful”

## These are the within-person correlations:
## These are the within-person p-values:
## These are the between-person correlations:
## These are the between-person p-values:
## These are the within-person correlations :
## These are the within-person p-values :
## These are the between-person correlations :
## These are the between-person p-values :
##                     Df Sum Sq Mean Sq F value   Pr(>F)    
## CareForAltruistic    1  15782   15782   29.62 6.56e-08 ***
## Residuals         1046 557373     533                     
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 4370 observations deleted due to missingness
## Tables of means
## Grand mean
##          
## 16.93798 
## 
##  CareForAltruistic 
##     care with self interest cared without self interest
##                       27.03                       15.45
## rep                  135.00                      913.00
##                     Df Sum Sq Mean Sq F value  Pr(>F)   
## CareForAltruistic    1   1924  1923.7   6.997 0.00829 **
## Residuals         1046 287578   274.9                   
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 4370 observations deleted due to missingness
## Tables of means
## Grand mean
##          
## 76.41221 
## 
##  CareForAltruistic 
##     care with self interest cared without self interest
##                       72.89                       76.93
## rep                  135.00                      913.00

Meta analysis

ivs <- c(“CareForOther_num”) dvs <- c(“HowUnhappy”, “WithoutEnergy”, “Distracted”, “Stressed”, “AtEase”, “Optimistic”, “Delighted”, “Satisfied”, “Grateful”)

What is the link between caring for others (=1 on careforother_num) and different indices of well-being?

Summary of Findings:

  • b: This represents the pooled or average effect across all people. Only two effects, delighted and grateful, were found significant.
  • SE and pval: These indicate the standard error and p-value of the pooled effect, respectively.
  • CI: Confidence intervals of the pooled effect. Intervals that do not overlap with 0 indicate significance. In this case, grateful is the only measure we can be totally confident in.
  • : Measures the level of heterogeneity among studies.
    • Less than 25%: Low inconsistency
    • 25% to 50%: Moderate inconsistency
    • 50% to 75%: High inconsistency
    • Over 75%: Very high inconsistency
  • Qe and Qep (p-value): These statistics test if variability among studies is significant. Given we had no apriori hypothesis about what would or would not be heterogenous, We focus only on Qep values that are p < .00001, also associated with I² above 50%.

In conclusion, for most variables, there is no significant pooled effect but there is significant variability, indicating that effects may exist for some people. Further investigation is needed to explore these effects.

##       variable     b   se pval ci_lb ci_ub    I2     QE QEp
##     HowUnhappy -0.09 0.05 0.07 -0.19  0.01 72.46 521.00   0
##  WithoutEnergy -0.05 0.04 0.22 -0.14  0.03 63.83 294.30   0
##     Distracted -0.04 0.04 0.38 -0.11  0.04 57.66 239.08   0
##       Stressed  0.05 0.04 0.14 -0.02  0.12 30.96 204.58   0
##         AtEase  0.05 0.04 0.20 -0.03  0.13 64.56 391.89   0
##     Optimistic  0.08 0.04 0.08 -0.01  0.16 59.70 316.27   0
##      Delighted  0.10 0.04 0.01  0.03  0.17 35.59 206.30   0
##      Satisfied  0.09 0.04 0.05  0.00  0.17 61.54 402.59   0
##       Grateful  0.07 0.04 0.06  0.00  0.14 37.73 212.26   0

Interclass correlations of key variables

ICC1 (Single-Raters Absolute Agreement)

ICC1 measures the proportion of total variance in the measurements that is attributable to the differences among subjects. It can be thought of as the consistency of scores within a particular participant when measured multiple times. Higher ICC1 values indicate that individual scores are more consistent for each participant, meaning that the variation within participants is low compared to the variation between participants. Conversely, lower ICC1 values indicate higher within-participant variance.

ICC2 (Average-Raters Agreement)

ICC2 assesses the reliability of the average of multiple measurements (e.g., multiple raters or multiple time points). It assumes that the measurements are a random sample from a larger population. Essentially, ICC2 reflects the reliability of the measure if all items nested within each participant (ID) were used to form an average. Higher ICC2 values indicate that the average scores are more consistent and reliable across different participants.

## Statistics within and between groups  
## Call: statsBy(data = new_data_frame, group = "ID")
## Intraclass Correlation 1 (Percentage of variance due to groups) 
##               ID CareForOther_num       HowUnhappy    WithoutEnergy 
##             1.00             0.14             0.35             0.45 
##       Distracted         Stressed           AtEase       Optimistic 
##             0.52             0.41             0.44             0.53 
##        Delighted        Satisfied         Grateful 
##             0.56             0.46             0.56 
## Intraclass Correlation 2 (Reliability of group differences) 
##               ID CareForOther_num       HowUnhappy    WithoutEnergy 
##             1.00             0.81             0.93             0.96 
##       Distracted         Stressed           AtEase       Optimistic 
##             0.97             0.95             0.95             0.97 
##        Delighted        Satisfied         Grateful 
##             0.97             0.96             0.97 
## eta^2 between groups  
## CareForOther_num.bg       HowUnhappy.bg    WithoutEnergy.bg       Distracted.bg 
##                0.17                0.37                0.47                0.53 
##         Stressed.bg           AtEase.bg       Optimistic.bg        Delighted.bg 
##                0.43                0.46                0.54                0.57 
##        Satisfied.bg         Grateful.bg 
##                0.48                0.58 
## 
## To see the correlations between and within groups, use the short=FALSE option in your print statement.
## Many results are not shown directly. To see specific objects select from the following list:
##  mean sd n F ICC1 ICC2 ci1 ci2 raw rbg ci.bg pbg rwg nw ci.wg pwg etabg etawg nwg nG Call

Cluster: How many factors?

Interpreting the Silhouette Plot

The silhouette plot produced by the fviz_silhouette function provides a graphical representation of how well each data point fits within its assigned cluster. Here is how to interpret the plot:

  1. Silhouette Width:
    • The silhouette width (represented on the x-axis) measures how similar each data point is to its own cluster compared to other clusters.
    • Values range from -1 to 1:
      • A value close to 1 indicates that the data point is well-matched to its own cluster and poorly matched to neighboring clusters.
      • A value close to 0 indicates that the data point is on or very close to the decision boundary between two neighboring clusters.
      • A negative value indicates that the data point might have been assigned to the wrong cluster.
  2. Clusters:
    • Each bar in the plot corresponds to a data point, grouped by their assigned cluster.
    • The clusters are separated by different colors and labeled on the y-axis.
  3. Average Silhouette Width:
    • The red dashed line represents the average silhouette width for all data points.
    • A higher average silhouette width indicates better-defined clusters.
  4. Optimal Number of Clusters:
    • The silhouette plot helps in determining the optimal number of clusters by visualizing the average silhouette width for different numbers of clusters.
    • Choose the number of clusters that maximizes the average silhouette width.

By analyzing the silhouette plot, we can assess the quality of clustering and identify if any data points have been misclassified.

## [1] 2
## Here are the silhouette scores for 2 to 10 clusters:
## [1] 0.27054196 0.16265428 0.06877118 0.08229655 0.09808085 0.10599966 0.11716224
## [8] 0.11480569 0.10581597
## We want to choose the number of clusters with the highest value.
## The highest silhouette score is 0.270542 for 2 clusters.
##   cluster size ave.sil.width
## 1       1   79          0.34
## 2       2   50          0.17

## 
##  1  2 
## 79 50

Cluster: interpretation

The table below represents the medoids of the clusters obtained from the Partitioning Around Medoids (PAM) clustering algorithm. Medoids are the representative points of each cluster, similar to centroids in k-means clustering, but medoids are actual data points from the dataset.

  • Medoids: The medoids are central data points that minimize the dissimilarity with other points in the same cluster. Each row represents a medoid for one of the clusters.
  • Rounding: The values have been rounded to two decimal places for simplicity and easier interpretation.

By analyzing the medoids, we can understand the characteristics of each cluster and how the representative points compare across different clusters.

Cluster 1 seems to involve caring having a positive effect oun well-being; cluster 2 suggests that carring had a negative effect

##   HowUnhappy WithoutEnergy AtEase Optimistic Satisfied
## 1      -0.20         -0.25   0.14       0.22      0.13
## 2       0.23          0.23  -0.20      -0.30     -0.23

Cluster: recognizing complex cases

We saw earlier that some people did not fit well into either cluster

A silhouette score of 0.10 for an individual data point can be interpreted as follows:

Poor Fit: The data point is poorly matched to its assigned cluster. Ambiguous Position: The data point is very close to the boundary between its assigned cluster and a neighboring cluster, indicating that it does not fit well with either. Possible Misclassification: The low score suggests that the data point might be misclassified and may belong to a different cluster. Low Confidence: There is low confidence in the clustering assignment for this data point, and it indicates that the clustering structure might not be optimal.

## # A tibble: 5 × 7
##   variable   cluster0_avg cluster0_se cluster1_avg cluster1_se cluster2_avg
##   <chr>             <dbl>       <dbl>        <dbl>       <dbl>        <dbl>
## 1 Unhappy         0.00635      0.143        -0.296      0.0470        0.485
## 2 NoEnergy        0.140        0.106        -0.270      0.0516        0.371
## 3 AtEase         -0.0688       0.0811        0.256      0.0442       -0.372
## 4 Optimistic      0.00865      0.0650        0.297      0.0426       -0.476
## 5 Satisfied      -0.203        0.130         0.337      0.0481       -0.386
##   cluster2_se
##         <dbl>
## 1      0.0935
## 2      0.115 
## 3      0.0864
## 4      0.0831
## 5      0.0748
## Number in each group is:
## 
##  0  1  2 
## 21 75 33

Cluster differences

##   Observed Expected WhoCaredFor Cluster3
## 1       38    41.72         Kin  HelpBad
## 2      167   163.28      NotKin  HelpBad
## 3      135   131.28         Kin HelpGood
## 4      510   513.72      NotKin HelpGood
## Generalized linear mixed model fit by maximum likelihood (Laplace
##   Approximation) [glmerMod]
##  Family: binomial  ( logit )
## Formula: kin ~ cluster3 + (1 | ID)
##    Data: long_data_with_cluster_noZero
## 
##      AIC      BIC   logLik deviance df.resid 
##    696.4    710.7   -345.2    690.4      847 
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -4.2529  0.1470  0.1877  0.3575  2.2765 
## 
## Random effects:
##  Groups Name        Variance Std.Dev.
##  ID     (Intercept) 4.326    2.08    
## Number of obs: 850, groups:  ID, 108
## 
## Fixed effects:
##                  Estimate Std. Error z value Pr(>|z|)    
## (Intercept)       2.42472    0.51380   4.719 2.37e-06 ***
## cluster3HelpGood  0.03696    0.56869   0.065    0.948    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr)
## clstr3HlpGd -0.779
## 
## Interpretation: The analysis found no significant differences between the groups in terms of helping kin versus non-kin.
## Generalized linear mixed model fit by maximum likelihood (Laplace
##   Approximation) [glmerMod]
##  Family: binomial  ( logit )
## Formula: CareForOther_num ~ cluster3 + (1 | ID)
##    Data: long_data_with_cluster_noZero
## 
##      AIC      BIC   logLik deviance df.resid 
##   3219.0   3236.7  -1606.5   3213.0     2753 
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -1.5120 -0.6583 -0.4259  0.9521  3.3221 
## 
## Random effects:
##  Groups Name        Variance Std.Dev.
##  ID     (Intercept) 0.7564   0.8697  
## Number of obs: 2756, groups:  ID, 108
## 
## Fixed effects:
##                  Estimate Std. Error z value Pr(>|z|)    
## (Intercept)       -1.1467     0.1787  -6.416  1.4e-10 ***
## cluster3HelpGood   0.3632     0.2115   1.718   0.0859 .  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr)
## clstr3HlpGd -0.843
## 
## Interpretation: According to the model, the HelpGood group tended to show a marginal tendency to help more than the other groups.
##           
##            Female Male
##   HelpBad      18   15
##   HelpGood     32   43
## Results of the Chi-Square Test:
## Chi-Squared Statistic: 0.87
## Degrees of Freedom: 1
## P-Value: 0.3519
## Conclusion: There are no significant gender differences. (p > 0.05)
## $avg_CareFeelings
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1     10    10.5   0.079  0.779
## Residuals   106  14028   132.3               
## 21 observations deleted due to missingness
## 
## $avg_CareVoluntary
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1     97   96.92   0.481  0.489
## Residuals   106  21355  201.46               
## 21 observations deleted due to missingness
## 
## $avg_CareObliged
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1    378   377.9   0.629   0.43
## Residuals   106  63715   601.1               
## 21 observations deleted due to missingness
## 
## $avg_CareGuilt
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1     40   39.91   0.147  0.703
## Residuals   106  28869  272.35               
## 21 observations deleted due to missingness
## 
## $avg_CareAvoidProblem
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1     27    26.7   0.082  0.775
## Residuals   106  34617   326.6               
## 21 observations deleted due to missingness
## 
## $avg_HowUnhappy
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1      8    7.81   0.032  0.858
## Residuals   106  25563  241.16               
## 21 observations deleted due to missingness
## 
## $avg_WithoutEnergy
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1      7     7.4   0.023   0.88
## Residuals   106  34344   324.0               
## 21 observations deleted due to missingness
## 
## $avg_Distracted
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1     30    30.2   0.086   0.77
## Residuals   106  37270   351.6               
## 21 observations deleted due to missingness
## 
## $avg_Stressed
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1     47   46.96   0.151  0.698
## Residuals   106  32867  310.06               
## 21 observations deleted due to missingness
## 
## $avg_AtEase
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1    403   402.6   1.736  0.191
## Residuals   106  24585   231.9               
## 21 observations deleted due to missingness
## 
## $avg_Optimistic
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1    347   347.3    1.21  0.274
## Residuals   106  30424   287.0               
## 21 observations deleted due to missingness
## 
## $avg_Delighted
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1    495   495.4   1.354  0.247
## Residuals   106  38794   366.0               
## 21 observations deleted due to missingness
## 
## $avg_Satisfied
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1    195   195.3   0.743  0.391
## Residuals   106  27859   262.8               
## 21 observations deleted due to missingness
## 
## $avg_Grateful
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1      1     0.8   0.002  0.965
## Residuals   106  44325   418.2               
## 21 observations deleted due to missingness
## 
## $avg_careforAltruistic
##              Df Sum Sq Mean Sq F value Pr(>F)
## cluster3      1  0.006 0.00622   0.147  0.702
## Residuals   106  4.492 0.04237               
## 21 observations deleted due to missingness
## 
## Conclusion: Based on the ANOVA analyses, there are no significant differences between groups in terms of well-being or motivation to care.

cohen’s d

Effect size differences between caring good and caring bad

##     Variable Cluster1_Mean Cluster2_Mean   Cohen_d
## 1    Unhappy        -0.296         0.485 -1.638918
## 2   NoEnergy        -0.270         0.371 -1.136592
## 3     AtEase         0.256        -0.372  1.416942
## 4 Optimistic         0.297        -0.476  1.811960
## 5  Satisfied         0.337        -0.386  1.708506
Cohen’s d for Cluster 1 vs Cluster 2
Variable Cluster1_Mean Cluster2_Mean Cohen_d
Unhappy -0.296 0.485 -1.638918
NoEnergy -0.270 0.371 -1.136592
AtEase 0.256 -0.372 1.416942
Optimistic 0.297 -0.476 1.811959
Satisfied 0.337 -0.386 1.708506