10. In Example 11.9 extend the varying slope model to all research inputs (lags 1 to 5 as well as the contemporary effect), as in (11.11). Following the McNab et al. (2004) strategy, it may be preferable to model the varying lag effects without a full 6-by-6 covariance structure, but first select lags where lag variation between firms is significant and then adopt a full covariance structure for that subset of effects. Does this model extension move the average deviance closer to the observation total of 1730? Another option is to allow firm-varying linear slopes (on time itself).