Regency EF
Persuading Investigators to Report Standardized Differences in Observational Studies (304077)
*Kyle Porter, The Ohio State UniversityKeywords: collaboration, communication, p-values, standardized differences, Table 1, observational studies
P-values are regularly used to compare groups in "Table 1" of observational studies. I will suggest that Table 1 p-values are inappropriate in observational studies and should be replaced by standardized differences. Studies that utilize propensity-score methods commonly use standardized differences to assess balance between groups; I will suggest extending the practice to all observational studies. Investigators are often resistant to not including p-values. There are several reasons not to use p-values for Table 1 comparisons; I will focus on the proposition that whether two groups are “different” should not be dependent on sample size. Investigators tend to agree with this but have never considered the implications. Standardized differences will be explained. Example tables will be presented for two scenarios having the same standardized differences for all variables but differing sample sizes and therefore different p-values. This example will illustrate to investigators that assessments based on p-values are misleading and group comparisons should be evaluated by standardized differences instead. The comparisons should be descriptive rather than inferential.