|
Abstract
|
Depression, coping, and attempted suicide are important topics of research in any society, and Saharia and Ghosh[1] studied these in Tezpur, Assam, during 2021-2022. Their study was discussed in eJournal Club India[2] in January 2024. Several important points emerged during the discussion. We present these in the hope that they will guide future researchers during data analysis and manuscript preparation in this and other fields of investigation. We do not present an exhaustive discussion of the paper; rather, we comment on specific points in Table 1 in which the authors[1] examine relationships between socio-demographic variables, severity of depression, and coping.
First, we discourage the routine, exhaustive, univariate examination of relationships between all independent variables and all dependent variables. Such analysis serves no useful purpose; it is a 'fishing expedition' that is widely regarded as a questionable research practice because it can produce false positive findings related to multiple hypothesis testing.[3] If at all such an analysis is conducted, it needs to be explicitly stated as exploratory, and appropriate corrections should be applied to correct for false positive errors. More appropriately, in Table 1, it would have sufficed to merely describe the socio-demographics of the sample so that readers could understand to what population the findings of the study may be generalized. On a related note, study protocols should ideally state primary and secondary objectives and inferential statistical procedures should be applied only to test hypotheses that were set with prior application of thought.
Second, the reflex tendency to categorize continuous variables is strongly discouraged because statistical power and precision are reduced.[4] Using examples from the study,[1] readers may note that age is a number and not a group. Education can be operationalized into years of study. Severity of depression is measured along a ratio scale and does not move in units of mild, moderate, severe, and very severe.
Third, when categorical variables are split into too many categories, meeting requirements for statistical testing becomes harder, statistical power is reduced, and interpretation of results becomes difficult. As an example, presenting socio-economic status in five categories was unnecessary because readers do not necessarily understand the data better; presenting it in lower, middle, and upper strata would have sufficed. Because of extreme sub-categorization, many cells for this variable were empty and the expected frequency requirement was violated for Chi-square testing.
Fourth, socio-economic status is an ordinal variable and severity of depression is a continuous variable. If indeed it was considered necessary to examine whether severity of depression varied across socio-economic strata, the Jonckheere-Terpstra test would have been apt; this is a test that many readers may be unaware of.
We hope that our observations will also be useful to journal reviewers who draft suggestions for improvements in manuscripts under review.
Language: en |