Wikipedia:Ambassadors/Research/Success factors

In this data project, we compared article quality results from each class to a set of course factors. At the moment, we focused specifically on Spring 2012 as a first priority, since it was the term most recently completed.

Because there was such a large difference between amount of improvement in existing articles and new articles, we split these up into two different categories so as not to distort the results in favor of all courses that created new articles rather than improving existing articles.

The results were not as distinct as we'd hoped. Of all of the course factors we measured, only three demonstrated any sort of statistical significance in predicting success of courses, and none of those do we feel strongly enough that the sample size is large enough to make it actionable. The statistically significant correlations were as follows:
 * Course Subject Area = d) humanities vs Mean (New Article Scores) is significant with p-value: 0.041964. In other words, courses in the humanities discipline who created new articles fared better than courses in other disciplines that created new articles. This result was not significant for improving existing articles, however.
 * Mean (Online Ambassador Number of Good Articles) vs Mean (Score Difference) is significant with p-value: 0.047198. In a surprising trend, classes had more quality improvement when the Online Ambassador working with the class had fewer Good Articles. This result seems counter to common sense, and the p-value is quite close to being insignificant, so we think we need to do further research on this point before enacting any sort of policy based on the result.
 * Number of Campus Ambassadors vs Mean (Score Difference) is significant with p-value: 0.029816. Similar to the previous result, this one surprised us, and we think it needs further research before we act upon it. We had run this number with an incomplete data set from Spring 2012 and in fact found the opposite, that the more Campus Ambassadors the course had, the better the results.

Ultimately, we believe this research needs more data sets before we can draw any meaningful conclusion from it in terms of a direction we would take. We are working on ways to get more quality reviews so we have a larger sample to draw from, and we will update this page as we have additional results.

Data analysis conducted by Luis Campos and Evan Rosen.