Covid-19 and altruism: a meta-analysis of dictator games

Abstract

Many growing studies have examined the impact of Covid-19 on altruism; the results, however, are divergent. This study synthesizes the rapidly expanding literature and performs a meta-analysis based on 24 dictator game studies reporting data collected after the start of the pandemic to examine whether Covid-19 framing and Covid-19-related recipients significantly impact altruism compared to neutral frame and non-Covid-19 recipients, respectively. Overall, the dictators donate about 42% of their endowment and depict relatively higher altruism when compared with other meta-analyses that used pre-pandemic studies. I also find that the Covid-19 and neutral frames lead to identical altruism. However, the dictators donate a higher fraction of endowment (about 6–9% higher) to the Covid-19-related recipients compared to those unrelated to Covid-19. These findings will provide helpful guidelines for future experiments focusing on the interplay of pandemic and altruism.

Appendices

Appendix A: Countries covered in the meta-analysis

No	Country	Studies	Estimates (K)
1	Austria	1	3 (2.44%)
2	China	4	7 (5.69%)
3	England	1	2 (1.63%)
4	France	1	1 (0.81%)
5	Germany	2	22 (17.89%)
6	Italy	1	1 (0.81%)
7	Netherlands	1	40 (32.52%)
8	Poland	1	2 (1.63%)
9	Russia	1	1 (0.81%)
10	South Africa	1	5 (4.07%)
11	Spain	2	3 (2.44%)
12	Sweden	1	2 (1.63%)
13	Taiwan	1	4 (3.25%)
14	USA	10	30 (24.39%)
Total	14	24	123

1. One study reports data from two countries (USA and Italy) and hence reported twice. Another study reports data from four countries (Germany, Poland, Sweden, USA) and hence reported four times

Appendix A1: Studies classified by framing and recipient type

ID	Study	Year	Neutral frame	Covid-19 framing	Covid-19 recipient	Other recipients
1	Adena and Harke	2022	Yes	Yes		Yes
2	Aksoy et al.	2021	Yes			Yes
3	Aksoy et al.	2023	Yes			Yes
4	Alsharawy et al.	2021	Yes			Yes
5	Blanco et al.	2021	Yes	Yes	Yes	Yes
6	Brañas-Garza et al.	2022	Yes			Yes
7	Chapkovski	2023	Yes			Yes
8	Chisadza	2023	Yes		Yes
9	Fanghella	2023	Yes			Yes
10	Fridman et al.	2022	Yes			Yes
11	Grimalda	2023	Yes			Yes
12	Grimalda et al.	2021	Yes		Yes
13	Guo et al.	2021	Yes	Yes		Yes
14	Hellmann et al.	2021	Yes		Yes	Yes
15	Lee et al.	2021	Yes	Yes		Yes
16	Li and Zheng	2023	Yes			Yes
17	Liebe et al.	2022	Yes			Yes
18	Livingston and Rasulmukhamedov	2023	Yes			Yes
19	Moon and VanEpps	2023	Yes			Yes
20	Romero-Rivas et al.	2021	Yes			Yes
21	Shachat et al.	2021	Yes			Yes
22	Sweijen et al.	2022	Yes		Yes	Yes
23	Wang	2021	Yes		Yes
24	Yue and Yang	2022	Yes		Yes

1. Studies 4, 20, 22, 23 and 24 use hypothetical decisions, while all other studies use incentivized decisions

Appendix B: Meta-synthesis based on the traditional models

Aggregation category	# of estimates (K)	Fixed effect modela	Random effects modela	Homogeneity test Q statistic [p-value]b
All observations	123	0.375*** (0.001)	0.419 *** (0.014)	36,178.68 *** [0.000]
Neutral frame	116	0.371*** (0.001)	0.417*** (0.014)	34,540.27 *** [0.000]
Covid-19 frame	7	0.528*** (0.006)	0.447*** (0.071)	947.83*** [0.000]
COVID unrelated Recipient	100	0.377*** (0.001)	0.401*** (0.015)	26,927.31 *** [0.000]
Covid-19 recipient	23	0.360*** (0.002)	0.498*** (0.032)	9209.95 *** [0.000]

1. Standard errors are in parentheses. ***p < 0.01
2. ^aNull Hypothesis = The synthesized effect is zero
3. ^bNull Hypothesis = The effect sizes are homogenous

Appendix C: MRA with random effects model

Estimator	Random effects
Regression #	[1]
Covid-19 frame	0.074
	(0.059)
Covid-19 recipient	0.072**
	(0.035)
Dictator type (base: students)	−0.200***
	(0.045)
Incentivized decisions	−0.067
	(0.048)
Gender (base: mixed)
Female	0.074
	(0.054)
Male	0.033
	(0.054)
Other	0.142
	(0.103)
Multiple donations	−0.007
	(0.053)
Matching subsidy	0.152***
	(0.051)
RPSP	0.084
	(0.066)
BRIS	0.084*
	(0.048)
Country (base: upper income)	−0.119***
	(0.042)
Year	0.010
	(0.020)
Published	−0.173**
	(0.078)
Constant	−18.662
	(39.447)
K	123
R-Squared (%)	37.40
Homogeneity test $\chi^{2}$	12,719.15
p-value	0.000

1. Standard errors are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10
2. Homogeneity test shows that random effects specification is appropriate

Appendix D: MRA with the two focus meta-independent variables combined

I created a new variable, “Framing + Recipient,” that takes on a value of 1 if either Covid-19 framing or Covid-19 related recipient is used and zero otherwise. The mean value for this combined variable is 0.23.

Estimator	Cluster-robust OLS	Cluster-robust WLS [df]	Cluster-robust WLS [1/SE]	Cluster-robust WLS [1/EST]	Cluster-robust random effects panel
Regression #	[1]	[2]	[3]	[4]	[5]
Framing + Recipient	0.085**	0.136**	0.119**	0.074	0.050***
	(0.035)	(0.057)	(0.044)	(0.079)	(0.018)
Dictator type (base: students)	−0.211***	−0.290***	−0.315***	−0.145**	−0.072
	(0.074)	(0.069)	(0.065)	(0.067)	(0.077)
Incentivized decisions	−0.063	0.062	0.003	0.011	−0.097
	(0.093)	(0.115)	(0.100)	(0.098)	(0.083)
Gender (base: mixed)
Female	0.076*	0.050	0.071	0.053	0.005
	(0.041)	(0.042)	(0.042)	(0.055)	(0.052)
Male	0.033	0.060	0.058	0.030	-0.038
	(0.051)	(0.045)	(0.049)	(0.064)	(0.061)
Other	0.150**	0.100	0.146**	0.110	0.075
	(0.061)	(0.075)	(0.063)	(0.086)	(0.055)
Multiple donations	−0.016	−0.074	−0.033	−0.113	−0.028
	(0.086)	(0.072)	(0.074)	(0.108)	(0.088)
Matching subsidy	0.159**	0.048	0.119	0.135	0.122*
	(0.059)	(0.082)	(0.070)	(0.082)	(0.070)
RPSP	0.088	0.105	0.137	0.132*	0.156**
	(0.099)	(0.091)	(0.085)	(0.065)	(0.071)
BRIS	0.083	0.004	0.111	0.006	0.122***
	(0.080)	(0.107)	(0.080)	(0.069)	(0.044)
Country (base: upper income)	−0.134**	−0.168**	−0.195***	−0.126*	−0.050
	(0.059)	(0.074)	(0.069)	(0.062)	(0.064)
Year	0.008	0.009	0.010	0.015	−0.015
	(0.024)	(0.049)	(0.028)	(0.029)	(0.033)
Published	−0.181	−0.147	−0.211	−0.278**	−0.174
	(0.128)	(0.117)	(0.125)	(0.122)	(0.107)
Constant	−15.279	−18.190	−20.364	−29.208	30.202
	(48.133)	(99.270)	(57.203)	(57.659)	(67.362)
K	123	123	123	123	123
R-squareda	0.455	0.418	0.520	0.388	0.114
Number of IDs					24
Breusch-Pagan test $\chi^{2}$					0.12
[p-value]b					[0.363]
Hausman test $\chi^{2}$					13.14
[p-value]b					[0.069]

1. Robust standard errors are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10
2. ^aR-squared within reported for regression number 5 (random effects regressions)
3. ^bBreusch-Pagan and Hausman tests (p < 0.05) indicate that random effects specification is appropriate

Appendix E: MRA for between protocol comparison

A between-protocol comparison is performed to examine the relative impact of the Covid-19 frame and Covid-19-related recipients on altruism. The base category for such a comparison consists of the neutral frame and non-Covid-19 recipients. The impact of the Covid-19 recipient is significant in three regressions and shows that the dictators donate about 6–10% more to such recipients than the control group, ceteris paribus.

Estimator	Cluster-robust OLS	Cluster-robust WLS [df]	Cluster-robust WLS [1/SE]	Cluster-robust WLS [1/EST]	Cluster-robust random effects panel
Regression #	[1]	[2]	[3]	[4]	[5]
Between protocol comparison (base: neutral frame + Non-Covid recipients)
Covid-19 Frame	0.122	0.277***	0.193*	0.103	0.023
	(0.095)	(0.085)	(0.110)	(0.083)	(0.057)
Covid-19 Recipient	0.077**	0.085	0.103**	0.053	0.056***
	(0.031)	(0.058)	(0.040)	(0.108)	(0.015)
Dictator type (base: students)	−0.210***	−0.288***	−0.311***	−0.144**	−0.069
	(0.073)	(0.069)	(0.064)	(0.068)	(0.077)
Incentivized decisions	−0.072	0.015	−0.013	−0.004	−0.096
	(0.091)	(0.114)	(0.095)	(0.096)	(0.084)
Gender (base: mixed)
Female	0.079*	0.062	0.076*	0.062	0.001
	(0.040)	(0.038)	(0.040)	(0.052)	(0.054)
Male	0.036	0.071*	0.063	0.038	−0.042
	(0.050)	(0.041)	(0.047)	(0.060)	(0.064)
Other	0.148**	0.109	0.144**	0.113	0.071
	(0.061)	(0.072)	(0.062)	(0.086)	(0.058)
Multiple donations	−0.017	−0.074	−0.031	−0.117	−0.025
	(0.086)	(0.072)	(0.074)	(0.110)	(0.088)
Matching subsidy	0.173***	0.112	0.145**	0.159	0.115*
	(0.058)	(0.083)	(0.064)	(0.097)	(0.069)
RPSP	0.094	0.128	0.144*	0.135**	0.155**
	(0.097)	(0.087)	(0.081)	(0.063)	(0.072)
BRIS	0.087	0.028	0.119	0.008	0.124***
	(0.077)	(0.104)	(0.073)	(0.067)	(0.044)
Country (base: upper income)	−0.139**	−0.164**	−0.203***	−0.127**	−0.045
	(0.066)	(0.068)	(0.070)	(0.061)	(0.069)
Year	0.011	0.011	0.014	0.019	−0.017
	(0.025)	(0.045)	(0.028)	(0.026)	(0.034)
Published	−0.188	−0.167	−0.219*	−0.291**	−0.165
	(0.131)	(0.118)	(0.126)	(0.128)	(0.110)
Constant	−20.841	−21.354	−28.469	−37.242	35.656
	(49.621)	(90.897)	(57.272)	(51.726)	(69.123)
K	123	123	123	123	123
R-squareda	0.457	0.452	0.525	0.391	0.121
Number of IDs					24
Breusch-Pagan Test $\chi^{2}$					0.08
[p-value]b					[0.386]
Hausman test $\chi^{2}$					13.21
[p-value]b					[0.105]

1. Robust standard errors are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10
2. ^aR-squared within reported for regression number 5 (random effects regressions)
3. ^bBreusch-Pagan and Hausman tests indicate that random effects specification is appropriate

Appendix F: Bayesian model averaging

Meta-independent variable	Coefficient	Standard error	PIP
Covid-19 frame	0.04	0.05	1.00
Covid-19 recipient	0.10	0.04	1.00
Dictator type	−0.14	0.03	1.00
Incentivized decisions	−0.00	0.01	0.09
Gender	−0.00	0.01	0.11
Multiple donations	0.01	0.03	0.22
Matching subsidy	0.06	0.07	0.56
RPSP	0.01	0.02	0.14
BRIS	0.01	0.02	0.13
Year	0.00	0.01	0.09
Published	−0.02	0.05	0.24
Country	−0.14	0.04	0.98
Intercept	−0.40	11.10	1.00
Observations	123
Model space	1024

1. PIP  = Posterior inclusion probability

Appendix G: MRA with robust meta-independent controls and published studies

Estimator	Cluster-robust OLS	Cluster-robust WLS [df]	Cluster-robust WLS [1/SE]	Cluster-robust WLS [1/EST]	Cluster-robust random effects panel
Regression #	[1]	[2]	[3]	[4]	[5]
Covid-19 frame	0.055	0.161**	0.100	0.090	0.003
	(0.049)	(0.065)	(0.061)	(0.056)	(0.043)
Covid-19 recipient	0.104***	0.067	0.099**	0.062	0.061***
	(0.032)	(0.063)	(0.039)	(0.080)	(0.012)
Dictator type (base: students)	−0.144***	−0.148***	−0.173***	−0.089*	−0.069
	(0.027)	(0.036)	(0.041)	(0.048)	(0.056)
Country (base: upper income)	−0.166***	−0.153***	−0.195***	−0.153***	−0.134***
	(0.038)	(0.051)	(0.030)	(0.044)	(0.048)
Constant	0.497***	0.518***	0.501***	0.471***	0.449***
	(0.016)	(0.017)	(0.009)	(0.041)	(0.057)
K	118	118	118	118	118
R-squared a	0.384	0.322	0.392	0.236	0.024
Number of IDs					22
Breusch-Pagan test $\chi^{2}$					0.74
[p-value]b					[0.195]
Hausman test $\chi^{2}$					6.55
[p-value]b					[0.088]

1. Robust standard errors are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10
2. ^aR-squared within reported for regression number 5 (random effects regressions)
3. ^bBreusch-Pagan and Hausman tests (p < 0.05) indicate that random effects specification is appropriate

Appendix H: Assessment of publication-selection bias by focus meta-independent variables

Figure 

Fig. 4

Funnel plot of fraction of endowment donated (sub-group analysis)

4 reports a funnel plot with fraction of endowment donated on the horizontal and standard errors on the vertical axis for two focus meta-independent variables. The plot for the neutral frame and non-Covid-19 related recipients appears to be symmetric and in the shape of an inverted funnel. However, for the Covid-19 frame and recipient, the shape does not appear to be inverted funnel; therefore, we perform the FAT-PET tests and summarize the findings in Table

Table 7 Summary of publication—selection bias

7.

The FAT test confirms publication-selection bias for the Covid-19 frame and the Covid-19 recipient protocols. I do not find conclusive evidence for publication-selection bias for the remaining two protocols. I follow the procedure of Stanley and Doucouliagos (2012) and Brada et al. (2021) to obtain publication-selection bias-adjusted estimates for the Covid-19 frame protocol. Specifically, we estimate the following equation:

$$t_{k} = \alpha_{0} SE_{k} + \alpha_{1} \left( {1/SE_{k} } \right) + \varepsilon_{k}$$

(3)

The null hypothesis $\alpha_{1} = 0$ tests whether a true effect exists. In the presence of a true effect, $\alpha_{1}$ is considered its publication-selection bias adjusted value. Stanley and Doucouliagos (2012) call this process as precision-effect estimate with standard error (PEESE). I use three estimators (unrestricted WLS, cluster-robust unrestricted WLS, and random effects panel ML) to estimate equation H and reject the null hypothesis ($\alpha_{1} = 0$) if at least two estimators provide significant value for $\alpha_{1}$. The publication-selection bias-adjusted estimate for the Covid-19 protocol is 0.597 and shows a donation rate of about 7% higher than that obtained from WAAP, reported in Table 3 in the main text. Similarly, publication—selection bias adjusted maximum value for the Covid-19 recipient protocol is 0.354, and closely matches WAAP reported in Table 3 in the main text.