biased-poll-replication-package

This repo contains the original data and the codes that reproduce the figures and results in the following paper:

A. Boukouras, W. Jennings, L. Li, Z. Maniadis,"Can Biased Polls Distort Electoral Results? Evidence from the Lab." European Journal of Political Economy, forthcoming.

• To make sense of the following explanations, you may want to have a quick look at the working paper here.

• You can also find the oTree source code for the experiment here.

original data

This folder contains all data from experiments E1, E2 and E3.

• E1_data and E2_data were collected at the Unversity of Southampton and Newcastle Business School between May and November 2018;
• E3_data was collected at the University of York in June 2019.

codes

The codes folder consists of two parts:

• descriptive: codes that produce the descriptive results, such as Figure 4, Figure 5 and results in the Welfare Effects and Individual Behaviour section. All codes are python.
• regression: codes that produce the regression results in Appendix B. python is used to clean the data, and stata is for regressions.

descriptive

1. running E1.py, E2.py, and E3.py gives you all the results in Results section. However, more work needs to be done to reproduce the exact figures.

   • For instance, running E1.py creates the following figure different from Figure 4 in the paper.

     

   • Also, after running E1.py, E2.py and E3.py, you will see the following .csv files in your working directory.

     

   • The csv files store the numbers in the figures. For instance, the following E1_win.csv is consistent with the bar graph above.

     

2. combine E1_win.csv, E2_win.csv and E3_win.csv, so you have a win.xlsx file looks like following:

   

   • combine all the *_vote.csv files, and impelement the same process for all *_pearon.csv files, and you will have a win.xlsx, a vote.xlsx and a pearson.xlsx that contain the results of all three experiments.
   • put all the these combined csv files in the same directory with win.py

3. run win.py, and it gives you the figures in Figure 4.

4. E1_additional_analysis.py,E2_additional_analysis.py and E3_additional_analysis.py produce results in Welfare Effects and Individual Behaviour section and the tables in Appendix D.

   • Note that code does not output the tables directly, and some straightforward calculations and copy&paste work are still needed.

regression

1. running E1_reg.py, E2_reg.py and E3_reg.py gives you All_E1.csv, All_E2.csv and All_E3.csv, and these are cleaned-up data for regressions.

   • Note that there is code for regressions at the end of the python scripts. However, I wasn't familiar with statsmodels package at the time, I switched to stata instead.

2. run cluster.do to generate the results in Appendix B Additonal Econometric Analysis。

   • The .do file tests six models and only "model 1" and "model 3" in the script are presented in the online appendix of the final version.

Notes: I apologise that the E1.py, E2.py and E3.py scripts are a bit messy. They are code drafts that also produce figures and results that, in the end, do not present in the final version of the paper. For the purpose of replicating the result of the final version, we commented out unnecessary code blocks. Should you have any questions, please email me at lunz3706@outlook.com.