In one of my studies, I need to assess the sample size. The study aims to predict the choice of participants with a continuous predictor and a categorical predictor (control vs. experimental). Here we hypothesize that the choice of the participants will be significantly predicted by the interaction between the predictors taken for the study. In this case, how to conduct a power analysis to find out the sample size required for the study.
Note: The alpha is set at 0.05, power/1-alpha/ beta is set at 0.80
The purpose of your research is to test the effect of the predictor interaction term. Thus you need only the sample size that gives you the power that you need for your test of that term which answers your question that is the interaction term. This is not as simple as it would be in ols regression. Some suggestions are given here: https://www.google.com/search?client=firefox-b-1-d&ei=2EeZX8GpMpC3tQa4z6fgAw&q=What+sample+size+for+a+logistic+regression+term+test&oq=What+sample+size+for+a+logistic+regression+term+test&gs_lcp=CgZwc3ktYWIQDFAAWABg-eIGaABwAXgAgAEAiAEAkgEAmAEAqgEHZ3dzLXdpeg&sclient=psy-ab&ved=0ahUKEwjB9bCEitfsAhWQW80KHbjnCTwQ4dUDCAw
Best, D. Booth
Thanks for your answer Dr. David Eugene Booth . I shall refer to the attached paper.
Hello, Balachandar Kaliappan David Eugene Booth
The purpose of your research is to test the effect of the predictor interaction term. Thus you need only the sample size that gives you the power that you need for your test of that term which answers your question that is the interaction term. This is not as simple as it would be in ols regression. Some suggestions are given here: https://www.google.com/search?client=firefox-b-1-d&ei=2EeZX8GpMpC3tQa4z6fgAw&q=What+sample+size+for+a+logistic+regression+term+test&oq=What+sample+size+for+a+logistic+regression+term+test&gs_lcp=CgZwc3ktYWIQDFAAWABg-eIGaABwAXgAgAEAiAEAkgEAmAEAqgEHZ3dzLXdpeg&sclient=psy-ab&ved=0ahUKEwjB9bCEitfsAhWQW80KHbjnCTwQ4dUDCAw
Getasew Amogne Aynalem I don't see a different answer from David.
For complex logistic regression models, it is advised to estimate power analysis using simulated data. I ran simulations using R and found that to detect a significant interaction effect, we need approximately around 500 sample. I ran the simulation to get a small effect size (cohen's d = 0.2).
Dear Balachandar, Kaliappen, It is just for recommending and sharing his vital and valuable answers as his justification and explanation was so impressive.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size