The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Error code: DatasetGenerationCastError
Exception: DatasetGenerationCastError
Message: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 7 new columns ({'denials', 'decisioned_n', 'group', 'lei', 'mix_predicted_rate', 'overlay_residual_pp', 'actual_denial_rate'}) and 3 missing columns ({'metric', 'value', 'definition'}).
This happened while the csv dataset builder was generating data using
hf://datasets/FinanceRateCalc/fha-door-effect-2025/overlay_residuals_2025.csv (at revision 1f51a94684e89987c005b36e4becebb47f1af23f), ['hf://datasets/FinanceRateCalc/fha-door-effect-2025@1f51a94684e89987c005b36e4becebb47f1af23f/door_effect_summary_2025.csv', 'hf://datasets/FinanceRateCalc/fha-door-effect-2025@1f51a94684e89987c005b36e4becebb47f1af23f/overlay_residuals_2025.csv']
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1848, in _prepare_split_single
writer.write_table(table)
~~~~~~~~~~~~~~~~~~^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 765, in write_table
self._write_table(pa_table, writer_batch_size=writer_batch_size)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 773, in _write_table
pa_table = table_cast(pa_table, self._schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
group: string
lei: string
decisioned_n: int64
denials: int64
actual_denial_rate: double
mix_predicted_rate: double
overlay_residual_pp: double
-- schema metadata --
pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 1139
to
{'metric': Value('string'), 'value': Value('float64'), 'definition': Value('string')}
because column names don't match
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1850, in _prepare_split_single
raise DatasetGenerationCastError.from_cast_error(
...<4 lines>...
)
datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 7 new columns ({'denials', 'decisioned_n', 'group', 'lei', 'mix_predicted_rate', 'overlay_residual_pp', 'actual_denial_rate'}) and 3 missing columns ({'metric', 'value', 'definition'}).
This happened while the csv dataset builder was generating data using
hf://datasets/FinanceRateCalc/fha-door-effect-2025/overlay_residuals_2025.csv (at revision 1f51a94684e89987c005b36e4becebb47f1af23f), ['hf://datasets/FinanceRateCalc/fha-door-effect-2025@1f51a94684e89987c005b36e4becebb47f1af23f/door_effect_summary_2025.csv', 'hf://datasets/FinanceRateCalc/fha-door-effect-2025@1f51a94684e89987c005b36e4becebb47f1af23f/overlay_residuals_2025.csv']
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
metric string | value float64 | definition string |
|---|---|---|
records_used | 859,090 | Decisioned FHA applications in model sample (actions 1,2,3; loan_type 2; complete covariates) |
raw_denial_rate | 0.2188 | Pooled denial rate on model sample |
mcfadden_r2_profile_only | 0.1712 | Model A: applicant/loan characteristics + state FE |
mcfadden_r2_with_lender | 0.276 | Model B: Model A + lender fixed effects (LEI) |
door_effect_share_of_explained | 0.3797 | (R2_B - R2_A) / R2_B — share of EXPLAINABLE variation associated with lender identity; associational, not causal |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
null | null | null |
The Door Effect (2025): Lender Identity and FHA Denial Outcomes
38% of the explainable variation in 2025 FHA mortgage-denial outcomes is associated with lender identity — not the applicant's file. McFadden pseudo-R² rises from 0.171 (applicant/loan characteristics only) to 0.276 when lender fixed effects are added, on 859,090 decisioned FHA applications from the complete public CFPB HMDA 2025 record.
Files
door_effect_summary_2025.csv— headline decomposition metrics with definitionsoverlay_residuals_2025.csv— top-15 strictest and top-15 most-lenient large lenders: actual vs mix-predicted denial rates; residual in percentage points (largest: +43.7pp)reason_fingerprints_context.json— denial-reason mix context for residual interpretation
Method (short)
Denominator rule: decisioned applications = action_taken ∈ {1,2,3}; denial = action 3; FHA = loan_type 2. Model A: logistic denial ~ income, loan amount, LTV band, DTI band, property type, state FE. Model B: Model A + lender (LEI) FE. Door-effect share = (R²_B − R²_A) / R²_B. Full methodology: https://financeratecalc.com/door-effect.html
Use boundary (read before use)
Associational, not causal. HMDA contains no credit scores; the residual means "not explained by observable federal-record characteristics." Never use this data for individual approval prediction, borrower profiling, or personalized lender recommendation. A lender's historical aggregate is not any person's probability. Machine-readable claim passport & contract: https://financeratecalc.com/claims/door-effect-38pct-2026.json
Provenance & citation
First published August 2026 by FinanceRateCalc (independent research, Ziya Yetiş). Cite: "FinanceRateCalc analysis of the public CFPB HMDA 2025 record; historical aggregate only." Related paper: SSRN 7156938 (AI-accuracy benchmark) · Working paper: The Door Effect (2026). License: CC BY 4.0.
- Downloads last month
- 14