8 Conclusion and Future Work

Overall, we have looked at some basic triple-double related figures of Russell Westbrook as well as analyzed the relationships between triple-double and some key basketball statistics. We have found that game result and opponent’s conference are useful categorical predictors of triple-double, whereas playing side is not. Additionally, quantitative variables like Westbrook’s playing time and plus/minus are effective in predicting whether Westbrook recorded double figures in three statistical categories in a basketball game. We also got a taste of some new visualization tools of mapping locations of observations onto the global map.

In the future, there are some additions we could make and some other topics we could explore. We could use a larger dataset, with not just data from Westbrook’s last 3 OKC seasons, but instead data throughout his career. Thus this would allow us to use randomization to select the cases, and hence would be able to generalize the results to a larger population of all Westbrook NBA games. In terms of model selection, we could try to determine the best subset of predictors for triple-double, or identify unusual points in regression.

Analysis-wise, looking at more complicated figures related to triple-doubles could be an option, data on topics like at what point of time in a game a triple-double is fulfilled, or streaks of triple-doubles could be collected and analyzed. In addition, instead of predicting triple-double outcomes, we could also predict other basketball statistics, most desirably, game results for OKC. One way would be to use factors such as points scored or shooting percentage to predict outcomes not just for individual games, but for an entire season, and then compare the predicted results with the actual results. More ambitiously, we could get some simulations involved and keep track of all sorts of metrics for Westbrook and his team.