Question

Study Design and Sample Size

The study is a non- randomized controlled trial including eight intervention and six control schools. The study assessed knowledge, attitudes and behaviors of the students three times over a period of eighteen months: March 2009 (Baseline), March 2010 (TI) and September 2010 (t20. We based the sample size calculation on the study objective of assessing whether or not the intervention influenced the time trend in condom use and recent history of sexual intercourse. Sample size calculations were conducted with Wald tests for odds ratio resulting from regression models with two binary variables (intervention/ control and TO/T1 OR T0/T2) and their interaction. For logistic regression models, a minimum of 1,241 observations are required to detect an adjusted odds ratio of 2 or more with 80% power under conservative assumptions of 30 % baseline prevalence of the outcome variable and no changes over time in the control group. For linear regression models, a minimum of 348observations are required to detect a small standardized effect size (Cohen’s d) of 0.3 with 80% power at the 0.05 significance level. Further, we assumed a design effect of 2, due to possibly strong correlation of repeated measurements from the same participant (TO/TI/T2), resulting in a minimum of 2,482 observations required from 1,241 participants. Anticipating a 25% loss t follow –up. We increased the target sample size to 1,655 participants at TO. Eventually for the research, the target sample was increased from 1,655 to 1950.

Question

The investigators described their sample size calculations, which are relatively complex. Did the investigators achieve the desired sample size? Do you think that the study findings are made stronger or weaker by the size of the sample? Explain your answer.

Answer 1

The proceeder to calculate the sample size is good and this is how the sample size is calculated. While calculating sample size we have to consider the prevalence of variable of interest, precision(margin of error), level of significance and power also. These are the basic things you need to look for before you calculate the sample size.  

Additionally, depending on the design and intra-class correlation the design effect comes into play. The 'design effect' compensates for the intra-class correlation in the population. So if there is an intra-class correlation in the population you will have multiply with design effect.

In the above sample size calculation, all these points have been covered thoroughly which yields a good sample size for the study. So it appears to be a good number of samples.

Moreover, a higher number of the sample make the estimate more precise. So making the size of the sample higher would make the study stronger. By a good number of samples, we can make the standard error of estimate lower which is again a good point,