Gordon Kuromi

It was not until about halfway through my B.Sc. program at Simon Fraser University that I took my first course in statistics. Until that point, the word "statistics" represented little more to me than what I had read in the sports pages of newspapers, even though I had already taken a number of math courses. But I found it to be an interesting subject, and took a few more courses in statistics before completing my degree, including a course on survey sampling that was the principal reason that I ended up working as a statistician at Statistics Canada.

Here at Statistics Canada, I work as a methodologist and the type of work that I do can be quite varied. The sampling designs and the methods used to produce estimates can differ substantially between surveys, and within a single survey itself, there are also many stages that require the attention of statisticians. As a result, the day-to-day work changes over time both between projects and within a given project. I have worked on aspects of the survey process, such as constructing a representative list (or frame) from which a sample can be drawn. At other times, I have worked on developing sampling designs to efficiently allocate sample sizes and select a probability sample from which to collect data. I have also spent time developing estimators to measure population variables of interest that make the most effective use of the sampling design and available auxiliary information, estimating the variance of survey estimates, and modeling variables to account for survey non-response and minimize bias that can potentially result as a consequence. I often have the opportunity to gain experience in different aspects of survey methodology, and each different survey invariably presents its own distinct challenges as well.

Over the past four years, I have worked on edit and imputation (E&I) for the Integrated Business Statistics Program (IBSP), a new framework for processing business surveys under a common framework with a focus on maximizing the use of generalized systems.  Imputation is a method of processing survey data that was never mentioned in any of my undergraduate statistics courses, and in fact it was only at Statistics Canada where I first heard the word “imputation” used.  Along with re-weighting, imputation is a method of "filling in" missing or non-response data in post-collection.  Although non-response is an inevitable aspect of any large-scale survey, the treatment of non-response was largely overlooked in the early theoretical development of sampling theory, and consequently there is little treatment of the topic in many textbooks on sampling theory. However over recent years, this has been changing, and more research has been devoted to this relevant topic. Imputation methods now include statistical modelling, historical comparisons possibly with modelled trend adjustments, and random donor methods.  Imputation offers certain advantages as well as disadvantages compared to re-weighting methods in the treatment of non-response.  Advantages include the preservation of sampling design weights for all units in the sample, and coherence of the data at the micro level.  Imputation is also the unique treatment method for non-response in the new Integrated Business Statistics Program.

In order for such methods to be used effectively and reliably, research has focused on minimizing the inevitable bias resulting from survey non-response, and accurately quantifying the additional variance in the resulting estimates due to imputation.  These aspects were key considerations in the re-design of the processing methodology for business surveys, which led to the new Integrated Business Statistics Program. Statistics Canada has developed a generalized system for edit and imputation that includes numerous methods of imputation built in-house using the SAS statistical software system.  This generalized system now forms a key component of the post-collection processing structure of the Integrated Business Statistics Program.

In my role on the project of integrating business surveys into this new framework, I collaborate with many other statisticians devoted to specific surveys in order to develop the most effective and efficient strategies for post-collection edit and imputation.  I also consult with statisticians responsible for the support and development of our generalized system for edit and imputation, and discuss needs that arise for the processing of business surveys entering this new framework.  I also frequently collaborate with partners that support the implementation of the Integrated Business Statistics Program, and meet with subject matter analysts in order to ensure that the methods implemented meet their needs as effectively as possible.  With so many surveys using the new framework, and with new technologies changing the way that survey data are collected, processed and analyzed, new challenges arise constantly.

There will always be new knowledge and expertise to gain, and I have enjoyed the opportunity to work with and learn from the many talented and experienced people who have been members of my projects, instructors of in-house statistical courses at Statistics Canada, professors of graduate university courses in statistics that I have taken, and some who have in fact been more than one of the above! I have also been pleased that I have had the opportunity to pursue further studies in statistics by completing a master's degree at Carleton University through part-time studies while working full-time at my position here (with support and accommodation from my employer ). I have always enjoyed being in a learning environment, so I appreciate working at a position in statistics with numerous learning opportunities, and where support is provided to take advantage of them.

• Connect with us
• Department of Statistics & Actuarial Science
Simon Fraser University
Room SC K10545
8888 University Drive
Burnaby, B.C.