Designs of small samples: Analysis by generalized least squares

The time series analysis (TSA) constitutes an appropriate procedure of anal ysis for interrupted time series designs (ITSD). The main disadvantage of t his analysis technique is that it requires a high number of observations wi th object of identifying the corresponding ARIMA model (autoregressive Inte grated Moving Averages). However, in applied behavioral investigation most of designs have small samples. As alternative to the TSA, it is Possible to appeal to the aproaches of generalized least squares (GLS). The main probl em for the aplication of GLS approach is the estimate of the residual varia ncie-covariance matrix. For this reason, in the present paper a new procedu re of GLS is studied, it is proposed as alternative solution to the analysi s of data of short time series with a single case and two phases (Arnau, en prensa). It is to apply the approach of ordinary least squares (OLS), tran sforming the original data and the design matrix by the square root or Chol esky factor of the inverse of the covariance matrix, under the assumption o f first order autoregressive stationary model (Fox, 1997). In this study is presented, by a Monte Carlo simulation using the MATLAB program (version 5 .2), the goodness of the proposed procedure.