Problems when interpreting research results using only p-value and sample size.


Book: 
Proceedings of the Seventh International Conference On Teaching Statistics (ICOTS-7), Salvador, Brazil.
Authors: 
Hoekstra, R., Kiers, H. A. L., Johnson, A., & Groenier, M.
Editors: 
Rossman, A., & Chance, B.
Category: 
Year: 
2006
Publisher: 
Voorburg, The Netherlands: International Statistical Institute.
URL: 
http://www.stat.auckland.ac.nz/~iase/publications/17/C437.pdf
Abstract: 

The following three probabilities seem crucial when interpreting data, especially in the behavioral sciences:1) the probability that an effect is present in the population, 2) the probability that a replication is significant; and 3) the probability that the effect for a single individual in the population is in the expected direction. In our study, we asked 51 subjects (university students and lecturers in psychology) to estimate these probabilities after reading a short description of a hypothetical experiment with as outcomes only p-value and sample size given. Large variations in estimated probabilities were found. However estimates of the probabilities tended to increase as a positive function of sample size for a fixed p-value. Simulation studies show that , assuming a uniform prior distribution for the parameter, this turns out to be incorrect for all three probabilities.

The CAUSE Research Group is supported in part by a member initiative grant from the American Statistical Association’s Section on Statistics and Data Science Education

register