Indian Journal of Psychological Medicine

COMMENTS ON PUBLISHED ARTICLE
Year
: 2019  |  Volume : 41  |  Issue : 2  |  Page : 197-

Multiple testing and protection against Type I error using P value correction: Application in cross-sectional study designs


Vikas Menon 
 Department of Psychiatry, Jawaharlal Institute of Post Graduate Medical Education and Research, Dhanvantri Nagar, Puducherry, India

Correspondence Address:
Dr. Vikas Menon
Department of Psychiatry, Jawaharlal Institute of Post Graduate Medical Education and Research, Dhanvantri Nagar, Puducherry - 605 006
India




How to cite this article:
Menon V. Multiple testing and protection against Type I error using P value correction: Application in cross-sectional study designs.Indian J Psychol Med 2019;41:197-197


How to cite this URL:
Menon V. Multiple testing and protection against Type I error using P value correction: Application in cross-sectional study designs. Indian J Psychol Med [serial online] 2019 [cited 2019 Jun 20 ];41:197-197
Available from: http://www.ijpm.info/text.asp?2019/41/2/197/252697


Full Text



Sir,

The issue of adjusting the P value for determining statistical significance for a family of related outcomes was highlighted by Andrade[1] recently using the example of a randomized controlled trial (RCT). This issue is also relevant, yet subtly different, in cross-sectional studies which are much more commonly conducted than RCTs across settings and, therefore, of interest to researchers.

To clarify, let us take the instance of a study that aims to examine the prevalence of depression in psoriasis compared with healthy controls and also looks at possible predictors of depression in the diseased using a multivariable procedure such as multiple linear regression (MLR). Let us assume that the independent variables being examined include coping, quality of life, and stress. Furthermore, coping and quality of life are measured by scales that yield four related domain scores, which are independently analyzed, whereas stress is a single score.

Now, there will be at least two sequential analyses from this research; the first one being the univariate analysis comparing the sociodemographic and clinical variables of interest (including the aforementioned variables) between groups and the second one being the regression analysis where variables have been selected on the basis of the results of the prior univariate analysis, as is the convention.

In this scenario, univariate testing requires P value correction only when examining coping and quality of life (but not for stress scores) and should be set at 0.05/4 (i.e., 0.0125) using Bonferroni method, or alternatively, this becomes the last P value for examination in Hochberg's procedure. This is because a correction is only required when different domains of the same construct (e.g., coping) are being examined. For the regression analysis, however, statistical correction of P value is not conventionally applied as this is a procedure that controls for multiple variables simultaneously. It is rare to see textbooks recommending adjustments for P value while testing partial regression coefficients in an MLR analysis. Researchers must balance the risk of type I error with being overly conservative and choose methods of statistical correction wisely. As a parting note, excessive dependence on P values as the basis for drawing study conclusions must be avoided.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.

References

1Andrade C. Multiple testing and protection against a type 1 (False positive) error using the bonferroni and hochberg corrections. Indian J Psychol Med 2019;41:99-100.