Strategies for construction hazard recognition
CONTRIBUTION TO STATISTICAL DATA HANDLING AND DETERMINATION OF
Download 2.75 Mb. View original pdf
|
| CONTRIBUTION TO STATISTICAL DATA HANDLING AND DETERMINATION OF EFFECT SIZE To accurately quantify and communicate the effects of an intervention on the variable of interest, rigorous and reliable data handling and analytical methods should be adopted. However, a review of literature in determining effect sizes for longitudinal studies indicate several methods, some of which have been heavily criticized by researchers. Therefore simplistic approaches such as the use of visual analysis techniques have been suggested (Cooper et al. 2006), however, more rigorous statistical handling is generally required for determining scientifically credible effects. Therefore, some researchers have contrasted the means between the outcome variable at the baseline and intervention phase. While this allows the comparison of effect sizes, the resolution of the data gathered is artificially reduced, thereby not accounting for the dynamic nature of the variable of interest (Polyhart wt al 2010). Also, more importantly, such methods do not capture effect sizes associated with changeover time, and the effect sizes inmost cases are skewed as the mean by itself does not capture the variability of performance. Among available more rigorous approaches, the most popular method for analyzing longitudinal time-series data is the autoregressive integrated moving average (ARIMA) modeling. While ARIMA models can efficiently and accurately quantify effect sizes, they generally require more 166 than 50 observations for building the model. Moreover, while the method facilitates valid statistical inferences, the results are not easily comprehensible. Therefore, in this study, we use time-series regression models presented by Huitema and Mckean (2000; 2007) and demonstrate how the method can be applied for construction research. The study clearly describes the metrics were developed, how measurements were undertaken in the field, and demonstrates the power of the time-series regression models to make valid and well supported inferences while adopting the multiple baseline approach. Download 2.75 Mb. Share with your friends:
|