To impute or exclude, what’s the impact?
Bibliographic record
Abstract
TPS 691: Methods of measurement, design and data analysis, Exhibition Hall, Ground floor, August 28, 2019, 3:00 PM - 4:30 PM Background: Missing data is persistently a thorn in the side of data science researchers. While there are various ways to deal with missing data, from basic to sophisticated, it is critical to understand the pattern and characteristics of your missing data (i.e. is it missing (completely) at random?). Most importantly, how does the missing impact the dependent variable? Aim: The purpose of this work is to assess the pattern of missing person-year data, particularly for the years leading up to the censoring event (e.g. death), and to examine the impact of excluding or imputing on survival model risk estimates. Methods: We use the 2001 Canadian Census Health and Environment Cohort (CanCHEC, N=3.1 million, 16-years follow-up). Residential postal codes reported on annual income tax filings were used to account for residential mobility among respondents and for exposure and area-based covariate assignment. Missing postal codes were assessed by proximity in years to censoring event (non-accidental mortality) and relationship to other covariates (education, income, etc.). Imputation took three forms, national annual mean, person-year mean, truncated postal code mean. We use Cox survival models to estimate hazard ratios of fine particulate matter (PM2.5) to assess the impact of exclusion versus imputation. Results: Four percent of person-years were missing postal codes, with slightly higher proportions in the years leading up to a mortality event. Imputation method had little impact on the overall PM2.5 mean (7.03 vs. 7.04 μg/m3); however, levels of PM2.5 were notably larger among subjects who died and were missing data compared to survivors (7.57 vs. 7.04 μg/m3) and tended to have lower socioeconomic characteristics. Preliminary results suggest that excluding person-years with the demonstrated pattern of missing can have large nullifying impact on the PM2.5 hazard ratio. Conclusion: Missing data can impact results. Discussion will focus on sensitivity tests to examine missing data.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.382 | 0.775 |
| Meta-epidemiology (narrow) | 0.002 | 0.002 |
| Meta-epidemiology (broad) | 0.004 | 0.005 |
| Bibliometrics | 0.003 | 0.008 |
| Science and technology studies | 0.003 | 0.012 |
| Scholarly communication | 0.010 | 0.013 |
| Open science | 0.005 | 0.007 |
| Research integrity | 0.007 | 0.014 |
| Insufficient payload (model declined to judge) | 0.017 | 0.006 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; the direct Gemma label and the distilled Codex classifier agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".