Spreadsheets for Analysis of Controlled Trials, Crossovers and Time Series.

Spreadsheets at this site for the analysis of controlled trials, post-only cross-overs and pre-post crossovers have been modified to allow adjustment for, and estimation of, the effect of two covariates (predictor variables). The user of the spreadsheet now has to specify a "custom" effect by inserting weighting factors to make any combination of each subject's repeated measurements. There is provision for estimating the custom effect as a trend in some or all of the repeated measurements representing a time series. Adjustment for a single covariate is obtained as previously, using the FORECAST function to fit a simple linear relationship between the covariate and the custom effect and thereby to predict the value of the custom effect at the mean or some other chosen value of the covariate. Adjustment for two covariates is achieved by predicting the value of the custom effect at chosen values of each predictor using the regression coefficients provided by the LINEST function. The standard error of the predicted value, which is required for inferential statistics, is not provided by LINEST but was derived from the standard errors of the regression coefficients by accounting for the correlation between the predictors. As before, individual responses to treatments are estimated by comparing variances of custom effects in experimental and reference (control) groups or treatments; in the case of post-only crossovers, two treatments are assumed to be repeats of a control or reference treatment, and these are now assigned to a separate spreadsheet to allow comparison of the variance of their change scores with the variance of change scores between reference and active treatments. The spreadsheets provide the usual analyses of the raw and log-transformed dependent variable, along with non-clinical and clinical magnitude-based inferences for smallest important effects defined by raw, percent, factor and/or standardized differences in the dependent variable. Values of the trial(s) or treatment(s) chosen to provide the standard deviation for standardizing are automatically selected as the values of one of the covariates, but these values can be replaced with those of another covariate. Accuracy of the computations in the spreadsheets was checked by analysing simulated datasets with the spreadsheets and with mixed modeling in the University Edition of the Statistical Analysis System (SAS Studio). [ABSTRACT FROM AUTHOR]