DG analysis for platform comparison
Peer, Brandt, Chapkovski
2023-11-13
Getting necessary libraries
Reading and cleaning data
Dictator’s decisions per platform
The main problem with the dependent variable (amount donated,
DG_4) is its underlying structure: the absolute majority
donates 50%, and an overwwhelming part do not donate more than 50%
(apart from mTurkers). We converted the DG_4 into 4-level ordinal
variable will report frequences of 50% and 0% donations plus run some
ordinal logistic regression on this categorical variable.
Converted results
| DG_4_ordinal | n |
|---|---|
| 0 | 316 |
| 0-4.99 | 625 |
| 5 | 1105 |
| 5+ | 471 |
modal choice (50%) distirbuted across platforms:
Share of 50-50 split by attention level
Experience with DG and its effect on DG decisions:
The naivete of participants by the first glance affects the
participant decision.
Naivete effect by platform:
Existing literature points that in real games more experienced users usually are more greedy (Brosig-Koch et al. 2017). In our case there the share of those choosing equal split is higher among unexperienced users for Besample, Toloka and MTurk platform and this intuition is not confirmed for Connect, CloudResearch and Prolific.
Gender difference
We drop ‘Other’ gender in further analysis because there are only 16 observations in this slot across all 6 platforms.
OLS regression results:
| Characteristic | Model 1 | Model 2 | Model 3 |
|---|---|---|---|
| OR (SE)1,2 | OR (SE)1,2 | OR (SE)1,2 | |
| sample | |||
| Besample | — | — | — |
| Connect | 0.86 (0.136) | 0.64 (0.153)** | 0.57 (0.180)** |
| CR | 0.46 (0.141)*** | 0.35 (0.157)*** | 0.31 (0.178)*** |
| MTurk | 0.09 (0.198)*** | 0.08 (0.203)*** | 0.08 (0.221)*** |
| Prolific | 1.01 (0.137) | 0.80 (0.151) | 0.73 (0.158)* |
| Toloka | 0.62 (0.134)*** | 0.64 (0.135)** | 0.63 (0.136)*** |
| DG_naive | |||
| No | — | — | |
| Yes | 0.89 (0.090) | 0.89 (0.091) | |
| attention.level | |||
| High | — | — | |
| Low | 0.58 (0.110)*** | 0.59 (0.111)*** | |
| Gender | |||
| Female | — | ||
| Male | 0.77 (0.089)** | ||
| Other | 2.98 (0.602) | ||
| Age | 1.01 (0.004)* | ||
| Income | 1.00 (0.021) | ||
| Education | 0.96 (0.027) | ||
| Null deviance | 3,373 | 3,373 | 3,373 |
| Null df | 2,516 | 2,516 | 2,516 |
| Log-likelihood | -1,549 | -1,535 | -1,525 |
| AIC | 3,111 | 3,087 | 3,076 |
| BIC | 3,146 | 3,133 | 3,152 |
| Deviance | 3,099 | 3,071 | 3,050 |
| Residual df | 2,511 | 2,509 | 2,504 |
| No. Obs. | 2,517 | 2,517 | 2,517 |
| 1 *p<0.05; **p<0.01; ***p<0.001 | |||
| 2 OR = Odds Ratio | |||
References:
Brosig-Koch, J., Riechmann, T. and Weimann, J., 2017. The dynamics of behavior in modified dictator games. PloS one, 12(4), p.e0176199.