Skip to content

Hazard estimates and MQ interpretation

This analysis compares the trained HCT separation-hazard coefficient with post-hoc logit estimates and translates MQ changes into changes in separation odds.

flowchart LR
    data[(Disaggregated regression data)]
    fit[[Fit three post-hoc logits]]
    cache[(Compact fits, metrics,<br/>and clustered covariance)]
    latex[/hazard_estimates.tex/]

    data --> fit --> cache --> latex

    classDef upstream fill:#e8eef7,stroke:#52739e,stroke-width:2px
    classDef input fill:#eee5f5,stroke:#76528c,stroke-width:2px
    classDef code fill:#f3f3f3,stroke:#666,stroke-width:2px
    classDef support fill:#fff0cc,stroke:#a66b00,stroke-width:2px
    classDef tableOutput fill:#e2f2e7,stroke:#3b7a4b,stroke-width:2px
    class data input
    class fit code
    class cache support
    class latex tableOutput
Inputs
Dataset Location
Disaggregated regression data (full_training) /data/<coverage>/analyses/full_training/disaggregated_regression_data/
Build
Component Location
Builder analyses/modules/hazard_estimates.R
Runtime hct-tower-analyses-cpu
Entry point analyses/pipelines/hazard_estimates.R / hazard-estimates
Output
Dataset File / location Grain and variables
Hazard-estimate results hazard_estimates.csv, hazard_estimates.tex, compact fits, metrics, and metadata in /artifacts/<coverage>/analyses/hazard_estimates/ One result per trained-model or post-hoc-logit specification

The analysis always uses full-training in-sample observations and includes calendar-year observations through 2024. It reports the trained model coefficient and three post-hoc logits:

  1. MQ with tenure fixed effects;
  2. MQ with tenure and calendar-year fixed effects;
  3. MQ and numerical seniority with tenure, calendar-year, worker-observable, and firm-observable controls.

Worker and firm fixed effects are not included. Post-hoc standard errors are clustered two ways by worker and firm. The table also reports pseudo-$R^2$, sample counts, and the odds change associated with +0.1 MQ and +1 SD MQ.

Command reference
Option Required / default Source-defined effect
--config PATH config/config_one_percent.yaml Analyses YAML config.
--shards ID[,ID,…] Unset Restricts the user-shard subset processed.
--rerun Off Refits all saved models instead of reusing existing persisted fits.
Option Required / default Source-defined effect
--task hazard-estimates Required Selects the CPU analyses task.
--config CONFIG.yaml, --shards ID[,ID,…], --rerun Config required; others off Passed to the R pipeline.
--partition, --time, --mem, --cpus, --job-name, --dependency, --dry-run Wrapper defaults Standard CPU overrides; default is 8 CPUs, 800 GB, and two days.
cd /hct-tower/analyses
Rscript --vanilla pipelines/hazard_estimates.R \
  --config config/config_one_percent.yaml
cd /hct-tower
slurm/submit.sh --task hazard-estimates \
  --config config_full.yaml

Valid cached models and metric sidecars are reused. Use --rerun only to refit all post-hoc logits. Existing caches made with the former 2023 ceiling cannot be reused for the expanded sample; pass --rerun on the first run after this change.

Sharded runs

Use --shards with comma-separated user-shard identifiers for a scoped, exploratory result. Outputs are isolated under shards/user_<ids>/.

cd /hct-tower/analyses
Rscript --vanilla pipelines/hazard_estimates.R \
  --config config/config_full.yaml --shards 00,01
cd /hct-tower
slurm/submit.sh --task hazard-estimates --config config_full.yaml \
  --shards 00,01

Each model sidecar stores the covariance matrix and table statistics, allowing the table to be reconstructed without rescanning the regression data.