Problems with assumptions used by “Determinants of Suicide Rates in Thailand “

[Jarunee Siengsanan-Lamont]

To be honest, I think the paper itself has issues that could make the models and conclusions invalid.
1. The dependent variable is the province’s suicide rate. Bangkok has an estimated population of 8 million, while Samut Songkhram has <200,000 people (Wikipedia.org). It is not surprising that Bangkok has the lowest suicide rate as the denominator(population) is enormous compared to other provinces. Adding an odd suicide case to the total suicide number of Samut Songkhram could result in a huge bias as the denominator is much smaller compared to adding one suicide case to Bangkok. The author did not mention any data transformation. I think the untransformed suicide rate already creates bias.

2. The author made a number of claims without reference. For example, the divorce rate, the author claimed that more educated people were likely to have a higher divorce rate. As for the consumption of alcohol, the author claimed that the rates were lower as people are more educated. The paper did not include a reference on the educational background of the provinces.

3. The author described that the prevalence of alcohol consumption rate was the rate of adults in the province consumed alcohol within one year of the questionnaire, which is meaningless. In my opinion, excessive drinking (alcoholic) and casual drinking habit could have a different impact on mental health and perhaps suicide rate. I think if the “alcohol” variable is derived from the “individuals who had consumed alcohol within one year” question, it is most likely be a confounder.

4. The residual plot (homoscedasticity) indicated that the linear regression assumption is invalid. Thus, linear regression is not suitable for the dataset.

5. The conclusion is also not justified based on the model outcome. The underage drinking variable is not significant (0.05<p<0.1) in the model2. However, the author recommended that the government should impose stricter law enforcement to prevent underage drinking, why?

[Chawarat Rotejanaprasert]

Great comments! I am aware of the issues appearing on the paper. Actually another type of modeling for counts and rates will be introduced next week and I also plan for each student to discuss the model assumptions and propose an alternative which are essentially what you just did! So perhaps it would be beneficial for the class to have informative discussions here for the week3 class project as well. Very well done!!! 😉

[John Robert Medina]

I would like to ask for a clarification. Based on my understanding of what was written in the paper, provincial data in 2011 was used. I am not certain if what the author had at hand was individual level data or an aggregated data per province. Because if it is aggregated data per province, the study design is not cross-sectional design but ecologic study design instead.

[Chawarat Rotejanaprasert]

The analysis was using secondary aggregate-level data. So the design would be retrospective and ecological.

[Author]

Posts