Eric Rancourt, SRMS Chair, Statistics Canada
As all of us statisticians are making sure to renew our ASA membership, serious consideration should be given to joining the Survey Research Methods Section (SRMS).
With the proliferation of data and data sources, one might think surveys are becoming less in demand or less important, but ensuring estimations and predictions are solidly anchored in a probabilistic approach that enables drawing solid inferential conclusions is needed more than ever.
Surveys are prime for anchoring non-random samples (administrative and other alternative files) into a framework that allows for conclusions beyond the data set at hand. As the use of machine learning and various more complex nonparametric (and parametric) models increases, this becomes key. Further, these techniques raise issues such as difficulty producing predictions for units that are outside of distribution (OOD) that are very much related to producing estimates in a context of large nonresponse rates. In a sense, survey statisticians have been working on developing solutions for these issues.
Finally, statisticians must combine data sets to produce valid integrated statistics. Again, survey statisticians have been working on this issue by combining the use of surveys and administrative data.
So, surveys are here to stay. They can provide a spine to numerous estimation and prediction challenges. They could even be designed to gather information specifically on OOD, thereby enabling greater possibilities of inference.
Of course, SRMS members also have much to learn from colleagues in other sections, so join us and let us benefit from each other.
Leave a Reply