**Analysis plan**
The main analysis will be done with mixed-effect logit model with binary indicator for incorrect choice of the side in each trial as the dependent variable and two random intercepts and one random slope, and several fixed effects. One random intercept captures variation in individual tendency to make errors in the dots task, second random intercept captures variation of incorrect answers across rounds, and the random slope captures variance in individual tendency to dishonesty. Fixed intercept in the model is proportional to the average likelihood of incorrect answers. Average dishonesty tendency is captured in this model as the main effect of a dummy variable indicating trials in which more dots is on the left side ("non-winning" side). Interaction of this variable with environmental attitude captures effect of environmental attitude on cheating. Interaction of non-winning side indicator and store condition indicator captures differences in cheating between different experimental conditions. Finally, a three-way interaction of the non-winning side indicator, environmental attitude and store condition indicator captures moderating effect of environmental attitude on cheating across experimental conditions. Apart from the above-mentioned predictors, order of the trial, and proportion of dots in each trial will be included as additional covariates in the model.
We are convinced that this model is appropriate to examine moral processes and moderating role of environmental attitude with the dots task data. However, due to the nature of the data (very low probability of cheating at the trial level) and the character of the model (complex interaction structure and three random effects), and based on simulations that we had run before data collection, we are concerned that it may not be possible to estimate the model with our dataset. If that is the case, we will use the second, Simplified Model.
Simplified Model essentially replicates individual-level data analysis as conducted by Mazar and Zhong (2010). The Simplified Model is an OLS model which has honesty score (e.g., Sharma, Mazar, Alter, & Ariely, 2014) as its dependent variable. Honesty score is an individual-specific difference between conditional probability of incorrectly identifying left and right side. We run this model on individual-level data (data collapsed over trials for each individual) with environmental attitude, experimental condition indicators and their interactions as independent variables. Main effects of these variables capture effect of environmental attitude and store conditions effects on honesty, whereas their interactions capture moderating effect of environmental attitude on moral processes linked to the store conditions.
Note that we will use two specification of experimental condition indicators (these alternative specifications are labelled *m1* and *m2* in the R script), which differ in how the contrast between the experimental conditions (i.e., two store conditions and the control group) are coded. Whereas *m1* uses control condition as the reference group for the experimental condition indicators, *m2* uses green store condition as the reference group.
The file “[dots task analysis][1]” contains an R script that we will use for analysis of data and testing of hypotheses through the two models. The file "[dots generation][2]" contains an R script used to generate the dots for the dots task.
[1]: https://osf.io/aqexd/
[2]: https://osf.io/2knav/
