Autoregressive time-series models were used to predict flight strip activities (i.e., LOOK, WRITE, and MANIP). With these types of models, complications arise when pooling across subjects to estimate the overall time-series regression model. Various methods of estimating the pooled regression model exist. The Seemingly Unrelated Regression (SUR) was chosen because it can be used when pooling across subjects or when lagged dependent measures exist in the time-series model (see Appendix for additional details). For pragmatic reasons, the forward selection procedure recommended by Gallant and Jorgenson (1979) was used with the restriction that individual variables not explaining at least four percent of the total variance were eliminated from the models.
First, the low- and high-complexity scenarios were divided into 90-second time intervals. The number of responses for an FPS activity during interval t
(Y t)
was predicted from the number of responses for the other P variables from that interval (Xpt ), and the number of responses from the previous interval
(Yt-1). That is, Y t=f(X1t, X2t, ..., Xpt
Yt-1)' SUR was used to develop separate models for each of the single and team conditions. Residuals were examined for each model and indicated that the data were well-fit by the model. Predictors are listed in order of contribution to the model. The models are divided into the categories, LOOK, MANIP, and WRITE. Each category will be discussed in turn.
The adjusted R2 represents the proportion of variance accounted for by the model, corrected for the number of predictors included in the model. For each model, the degrees of freedom, adjusted R2, X2, and corresponding p-value are reported. No reliable predictors were found for those models containing only an intercept.
LOOK. As shown in Table 2, looking was not predictable in either the single or team situations. Although three out of the four p-values were significant, indicating above chance predictability, only the high-complexity RLOOK model accounted for over 4% of the variance. This model indicates that when the traffic situation is complex, the R-side controller is likely to look at the strips when communicating with the D-side controller. Also, if the R-side has looked at the strips once, he or she is likely to look again. This is the only LOOK model that accounts for over ten percent of the variance. Because no general conclusions can be reached by imposing meaning on one model, the LOOK models will not be discussed further. As indicated previously, DLOOK is not included in the table because it indicated a look to the radar scope and is therefore not considered a flight strip activity.
本人是门外汉,能否帮一下啊!!