Data analysis that fails to account for independent groups defined by a subject characteristic (e.g., sex) or by a design characteristic (e.g., treatment order) can result in bias, confounding, and loss of precision in the outcome. Combining the outcomes from separate analyses of the groups is a robust approach to the problem that is easily achieved with the spreadsheet presented here. Differences in the outcome between groups represent the effect of the characteristic on the outcome, while the mean of the outcomes represents the outcome adjusted appropriately for the characteristic. The spreadsheet calculates confidence limits for the differences and for the mean from the confidence limits for the outcome in each group. It also presents magnitude-based inferences for the differences and mean. There are separate cells in the spreadsheet for outcomes represented by means or other normally distributed statistics, relative rates (risk, odds and hazard ratios) or other log-normally distributed statistics, and correlation coefficients.
[1]
Will G. Hopkins,et al.
A spreadsheet for deriving a confidence interval, mechanistic inference and clinical inference from a P value
,
2007
.
[2]
R. Fisher.
014: On the "Probable Error" of a Coefficient of Correlation Deduced from a Small Sample.
,
1921
.
[3]
F. E. Satterthwaite.
An approximate distribution of estimates of variance components.
,
1946,
Biometrics.
[4]
Alan M. Batterham,et al.
Spreadsheets for analysis of controlled trials, with adjustment for a subject characteristic
,
2006
.
[5]
Alan M Batterham,et al.
Making meaningful inferences about magnitudes.
,
2006,
International journal of sports physiology and performance.