Dear Su Rou Low

The number of covariates that are admissible (i.e., possible risk factors, confounding or interaction variables) is determined by the number of observations or sample size. I recommend just considering variables that are potential risk factors for the desired outcome.

To put it another way, compute the measure of association both before and after controlling for a potential confounding factor. Confounding was observed if the discrepancy between the two measures of association was 10% or more. If it is less than 10%, there was very little, if any, confusing.

The level of confounding is determined as the percent difference between the crude and adjusted measures of association (for either a risk ratio or an odds ratio): If the percentage difference is 10% or larger, we infer that confounding occurred.

Aside from the study's objective and population size, three factors are typically required to establish the proper sample size: the level of precision, the amount of confidence or risk, and the degree of variability in the variables being studied (Miaoulis and Michener, 1976)

Best Regards

Dr. Fatemeh Khozaei

Su Rou Low The free sample size calculation & power analysis software G*Power has options for multiple regression, including R-square change (increase) that may be helpful:

https://www.psychologie.hhu.de/fileadmin/redaktion/Fakultaeten/Mathematisch-Naturwissenschaftliche_Fakultaet/Psychologie/AAP/gpower/GPower31-BRM-Paper.pdf

A more comprehensive way to examine the sample size requirement is by means of a Monte Carlo simulation. In such a simulation, you can simultaneously specify all hypothesized effects (regression coefficients, correlations), including the effects of the confounders and the correlations among all of the predictors (including confounders) and find out about power for different sample sizes. See:

Article How to Use a Monte Carlo Study to Decide on Sample Size and ...

https://www.youtube.com/watch?v=hjAJSPGMQ4E

Fatemeh Khozaei

Dear Dr. Fatemeh Khozaei

Thank you for your response.

I already have my confounders and sample size (without taking confounders into consideration) determined, however, on top of that, I was suggested to take confounders into consideration since I had my sample size estimated with four factors (1) the probability of falsely rejecting the null hypothesis when it is true (p>0.05), (2) desired statistical power, (3) anticipated effect size and (4) number of predictors. Given it has included the main variable of my study as predictors, confounders (possible risk factors) are also suggested to be taken into consideration when estimating my sample size.

Therefore, I am unsure of how to include the confounders when calculating my sample size.

Regards,

Su Rou

Christian Geiser

Dear Dr. Christian Geiser

Thank you for your recommendation in using Monte Carlo Simulation in estimating my sample size. As it is recommended with the use of SEM, I am unsure if it is applicable if I plan to use process macro? May I know more about it? Thank you.

Su Rou Low Linear regression is a special case of SEM. In other words, you can specify a multiple regression model as an SEM and simulate the parameter values in a Monte Carlo simulation. I offer a free workshop on how to apply simulations to sample size planning using Mplus. See:

https://www.goquantfish.com/courses/sample-size-planning-in-mplus

The workshop uses path analysis as an example, which is a generalization/extension of linear regression.