1. NLSCY – Non-response

2. Non-response There are various reasons why there is non-response to a survey  Some related to the survey process Timing Poor frame information Interviewer or field errors  Some related to circumstances Weather Language issues Difficulty in tracing individuals  Others related to respondents Unwillingness to participate Unable to participate

3. Variety of non-reponse Partial non-response (item)  Some individual questions were not answered  Some individual questions were not asked Partial non-response (component)  The NLSCY is sectioned into different groups of questions related to various topics; an entire section may be missing.

4. Variety of non-reponse Total non-response  No information is collected or  Insufficient information is collected Wave non-response  Where information about a respondent is available but not for every cycle of the survey due to total non-response in a given cycle  Follow-up strategy differs upon the cycle of introduction in the survey

5. Dealing with TOTAL non-response Total non-response is measured and corrected in the NLSCY at Statistics Canada  Significant variables which affect total non- response are identified  All the weights are adjusted to correct for the total non-response However, wave non-response is still present for longitudinal analysis

6. Adjustments for total non-response

7. Adjustments (cont’d)

9. Dealing with PARTIAL non-response Missing data for variables related to income are imputed at Statistics Canada Missing data for the other variables are identified as:  Refusal  Don’t know  Not stated Note: These are different from:  Not Applicable.

10. Dealing with partial non-response Issues Either analyzing the entire dataset  Where a significant amount of information is missing about a variable of interest  Or where many variables of interest have missing data and only a minority of records have all the pieces of information Limiting your analysis to a subset of the population where you have reported values  How do you make inferences to the larger population (question of what the weighted estimates refer to)

11. The Partition Family Dealing with non-response in partitioned datasets Missing an entire component Missing partial information Missing a cycle or wave of data

12. How important is it ? Maybe non-response is random. Maybe it's negligible Maybe it can be explained away Maybe I can get away with it

13. What are your options Report missing data as a value Ignore missing data (limit your analysis to reported data only) Correct for the missing data  By re-weighting  With imputation  model non-response information

14. Get to know your non-respondents When you have significant non-response  You need to assess non-response  It becomes your first variable of interest It’s an analysis like any other analysis you will do Otherwise it casts doubt over every finding

15. Example of ignoring non respondents in your analysis Based on the whole population… We know that the missing information relates to this sub population… Based on those who reported, we find that …. Inferences are now about a sub- population only. Relies on a good description of non-respondents | respondents

16. Example of imputing for non respondents in your analysis Based on the whole population… We know that the missing information relates to this sub population… We compensated for this non- response by doing the following, and based on this process, we find that …. Inferences are now about the population. Relies on a good description of your imputation methodology

17. Reweighting to compensate for non- response Same principle as imputation  Works when doing a whole components of missing values  Very messy in the Swiss-cheese type of non- response Composite methodology of imputation to adjust for local areas of non-response and re-weighting for broad areas where many variables (entire component) are missing.

18. Impact of non-response on variance Total non-response:  Variance will increase

19. Impact of non-response on variance Partial non-response: When imputing for non-response:  Treating imputed values for reported values may lead to an underestimation of the real variance (mean imputation is probably the worst case). When re-weighting:  How to use the Bootstrap weights?

20. Conclusion Make sure you always assess the impact of non-response when doing analysis