Data Bridging: Statistical Methods for Integrating Results across Studies

Abstract

A procedure is described for integrating data sets from several studies based on a sequential research strategy. A data bridging technique is used so that the results of three previous studies can be combined to build integrated second-order empirical models. The previous studies investigated 10 independent variables, but 16 uninvestigated interactions from the complete set of items in a second-order model were not investigated in these studies. To select the best data points for examining these uninvestigated interactions, the maximization of |X'X| criterion was used. A computer program was developed to calculate the determinant value of X'X matrix for each candidate data point. Each selected data point was then analyzed to determine multicollinearity of the uninvestigated interactions. From these analyses, six additional data points were selected to examine the 16 uninvestigated interactions. Results from six additional data points along with the previous data sets were combined to build second-order empirical models using polynomial regression. The empirical models were then analyzed to choose the best models in terms of statistical properties such as variances of coefficients and prediction variances of the models. This procedure is discussed as a method for integrating data collected through sequential experimentation into an empirical model describing the functional relationships among independent variables. This approach increases the generalizability of data used in the design and evaluation of human factors interfaces which involve a large number of factors.