Stepwise Variable Selection for Loglinear Mixture in Record Linkage
Bibliographic record
Abstract
A model building strategy is proposed to improve the probabilistic match in record linkage with focus on the loglinear mixture model of two components, each for the matched and unmatched pairs respectively. In reality, the comparison attributes (i.e., covariates) often interact each other, leading to more or less interactions in the loglinear models for both matched and unmatched pairs. However, the interactions patterns often are not the same for both components. Particularly, because the number of matched pairs is very small comparing with that of unmatched pairs in a real case, the model for matched pairs can not be fitted with the same higher order interactions as that for the unmatched pairs. The proposed strategy attempts to avoid both underfitting and overfitting due to subjective model specification for the data. Unlike the subjective specification, this strategy is data-driven. Starting from the situation of no interaction, we add interactions sequentially in two loglinear components using the forward selection approach. To this end, we define the alternatively climbing pathways through mixture families of two components with higher order interactions. The mixture models expanded along a pathway are nested successively, thus, conventional tests used for nested models can be applied. Regarding parameter estimation for the mixture, a simplified method (including the choice of initial values of parameters) for the EM algorithm is developed, which facilitates the mixture model fitting using existing packages and functions in sophisticated statistical software such as R. Simulation study has then been conducted for various situations to assess the model selection approach, and comparison of these selected models with the naive model assuming field independence has been made. We apply this strategy to the record linkage case study in SSC 2006 and have identified interactions among certain comparison attributes for both matched and unmatched pairs, these interaction patterns are not always the same for both matched and unmatched pairs.
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 | 0.020 | 0.031 |
| Meta-epidemiology (narrow) | 0.002 | 0.002 |
| Meta-epidemiology (broad) | 0.003 | 0.005 |
| Bibliometrics | 0.005 | 0.005 |
| Science and technology studies | 0.003 | 0.002 |
| Scholarly communication | 0.003 | 0.004 |
| Open science | 0.007 | 0.006 |
| Research integrity | 0.002 | 0.004 |
| Insufficient payload (model declined to judge) | 0.005 | 0.002 |
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; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
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".