In February, I predicted the results of the 2014 NHL regular season based on the results of the Men’s Olympic hockey rankings. Now that the season is over, the projections can be compared to the results:
A few of my predictions were well off (e.g., Vancouver, Buffalo, Detroit, Colorado, and Boston), but I still wanted to determine if my Olympics predictions were correlated to a statistically significant level with NHL season results.
The first step is to formulate my hypothesis (which I somewhat alluded to in my previous post). The null hypothesis—which I hope to reject—is that my Olympic predictions have no correlation to the NHL season results (they are independent). The alternative hypothesis is that there is a correlation.
The next step is to find an appropriate test statistic. Since these lists are both rankings (i.e., ordinal data), either Spearman’s rho (ρ) or Kendall’s Tau (τ) would be a good choice (they can be used to measure rank association, or the similarity of ordered rank data). I will use both for robustness. They should tell us if the two variables (predicted and actual) are statistically dependent. The null hypothesis, that the two are independent, would yield a value of zero for both measures. An ideal correlation would give a value of +1 (perfect positive correlation).
Here is perfect correlation:
Here are my results:
At a glance, the correlation does not appear strong. But is it statistically significant?
Both Kendall’s Tau and Spearman’s rho reject the null hypothesis of independence at an alpha of 0.1%. Kendall’s Tau is 0.46 and Spearman’s rho is 0.64—both suggest a statistically significant correlation between the rank I predicted based on the success of players’ teams in the Olympics and the NHL season results.
If not for the five “outlier” teams I mentioned earlier, the results become extremely promising (Spearman’s rho > 0.86). However, manipulating the data after the fact is questionable without a good reason. What would be better is to find the reason why those five teams did so much better/worse than my Olympic prediction projected.
Posts in this series: