androdri1 / dceillicitcolombia

Replication kit for: Rodríguez-Lesmes, P., Góngora, P., Mentzakis, E., Buckley, N., Gallego, J. M., Guindon, E., Martinez, J. P. Paraje, G. (2024). Would Plain Packaging and Health Warning Labels Reduce Smoking in the Presence of Informal Markets? A Choice Experiment in Colombia. Social Science & Medicine, 117069

This repo includes the code required to replicate the results in this paper. The data needed is available at: https://doi.org/10.34848/DLYZZU . Such data repository also includes the LimeSurvey questionnaire editable to facilitate reproducibility.

Any questions, please contact: Paul Rodríguez-Lesmes, Universidad del Rosario, [email protected] - General questions and Limesurvey Emmanouil Mentzakis, City, University of London, [email protected] - DCE Design Juan Pablo Martínez, Universidad del Rosario, [email protected] - Code (Stata and Mathematica)

Main Programs

Data process

1_data_process.do transforms the raw data to a format that can be used for descriptive statistics and running the regressions, as well as the value of information (VOI) computations. Prices in COP are transformed to USD using a fixed exchange rate of 3.800COP per USD.

Descriptive statistics

Descriptive statistics of the covariates at the respondent’s ages and blocks are built, which correspond to Tables 2 of the main paper and Tables 1 through 3 of Appendix A. 2_descriptives_tables.do generates these tables before minor editing.

Results – Multinomial logit

Multinomial logit models are run on 3_reg_tables.do where main results are built as well as supplementary material. Specifically, Tables 3 and 4 of the paper and Tables 4 and 5 of Appendix A.

Appendix A Programs

Heterogeneities

Differential results depending on age, income and consumption patterns are presented. These results are store in table 6 of Appendix A and the script that computes these heterogeneities for each design is A1_heterogeneities.do.

Latent class WTP

Latent class logit models are employed for computing the WTP and see whether there could be differences due to underlying characteristics of the respondents. These results are computed with A2_lclogit_wtp_w_bootstrap.do, which in turns employs two additional functions stored in func_lclogit_boostrap_program.do and func_lclogit_wtp_compute.do.

Latent class coefficients and characteristics per class

Coefficients and descriptives for each class depending on the resulting class are presented in table 9 in Appendix A and computed with A3_latent_class_tables.do.

Robustness

Main results robustness checks are checked depending on the time spent responding, keeping only non-dominated options, and dropping the 7th question. The results reported on table 10 of Appendix A are computed with A4_robustness.do.

Appendix B Programs

VOI data set up

Value of information calculations rely upon the computation of a price different to the one respondents face, so the database for finding this variable is generated in B_1_voi_data_process.do. The last price consumers paid when smoking is assigned in case respondents had a WTP equal to zero.

P^'calculations

B_2_p_p_function.nb is a wolfram mathematica notebook that executes the program for finding the specific price for each respondent. Its inputs correspond to the data file generated in the previous step. Given the necessity of having the specific program, the data is already provided in the raw_data folder.

VOI calculations

Having all the variables for computing the VOI and fixing the price and revenue elasticities, B_3_voi_calculations.do perform the computations of the final values. Moreover, the original data with demographics data is added for later generating tables.

VOI tables

B_4_voi_calculations.do generates both panels of the table for VOI results.