Search

Interrupted time series analysis involves repeated observations on a single unit both preceding and following an intervention. Recently developed statistical procedures permit inferential statements about the degree and pattern of the intervention effect. Five different software packages appropriate for interrupted time series analysis were evaluated: TSX, GENTS, BMDP, SAS, and ITSE. Simulated data was generated from 44 types of series representing eleven different ARIMA models, two types of intervention, and two different series' lengths. The five programs are compared with respect to the accuracy of estimation of the error variance, pre-intervention level, post-intervention level, and slope estimates (where appropriate). Three programs produced generally satisfactory results (TSX, GENTS, and SAS), one was generally inaccurate across a wide range of models (ITSE), and one produced some inaccuracies and occasionally failed to complete the analysis (BMDP). [ABSTRACT FROM AUTHOR]

Computer generated data representative of 26 ARIMA models was used to compare the results of interrupted time-series analysis using: (1) the known model identification, (2) an assumed (1, 0, 0) model, and (3) an assumed (3, 0, 0) model as an approximation to the General Transformation approach. Ten samples each of 26 ARIMA processes were generated representing ARIMA (0, 0, 0), (0, 0, 1), (1, 0, 0), (0, 1, 0), and (0, 1, 1) models with varying degrees of dependency. Series were generated with both 40 points and 100 points, with an immediate and constant intervention effect of magnitude one (occurring at the midpoint), and with a random error component variance of magnitude one. The results of the three analyses are compared to the criterion values and to each other for the point estimates of: l) the minimum residual error variance; 2) the pre-intervention level; and 3) the post-intervention change in level. The results for models which included a difference parameter were inaccurate for all three approaches. All three analyses provided reasonable and equivalent results for the remaining or non-differencing models for series with both 40 and 100 points The findings strongly suggest that the problematic model identification process might be eliminated entirely and replaced with the assumed (1, 0, 0) approach or the General Transformation approach. [ABSTRACT FROM AUTHOR]