Robustness

Each headline estimate is re-computed under every combination of reasonable analysis choices. A claim is reported as supported only if its sign survives all of them.

Every specification's trend estimateOne dot per specification; tick marks the median. The right margin gives the share of specifications with a positive trend.

Table view

Estimand	Specifications	Positive	Median	Min	Max
H1 · all people	900	899	4.39	−0.27	5.84
H2 authority · A	864	864	6.15	1.51	7.47
H2 authority · B	900	894	5.88	−2.03	9.23
H2 authority · A (app. sample)	144	144	6.59	2.25	7.77
H2 authority · B (app. sample)	144	144	6.44	2.98	9.85
H2 authority · C at 0.3	144	144	3.91	1.36	7.04
H2 authority · C at 0.5	144	144	3.76	1.23	7.44
H2 authority · C at 0.7	144	144	3.82	1.13	7.93
H2 authority · D at 0.3	144	144	5.21	2.22	7.77
H2 authority · D at 0.5	144	144	4.93	2.00	7.81
H2 authority · D at 0.7	144	144	4.60	1.79	7.70
Civic roles · B	864	864	6.36	2.63	7.59

The grid

Gender rule: honorific only; + pronoun rule (primary); + gendered nouns. People: NER-detected only (primary); + honorific-pattern detector. Duplicates: keep; drop near-duplicates (MinHash Jaccard ≥ .8). OCR: all; legible articles with dictionary rate ≥ .75. Newspapers: all; excluding the Evening Star; capped at 500 articles per paper-year. Time bins: year; 5-year; decade. Content: all; news only. Entity resolution: couple clusters such as “Capt. Bissell … Mrs. Bissell” set to UNKNOWN (primary; decision D17, added after first results); kept as merged. Role method and classifier threshold (0.3/0.5/0.7) for role estimands. A reduced grid adds topic + newspaper fixed effects.

By gender rule

Tier	Estimand	Cells	Positive cells	Sig. positive	Sig. negative	Median slope	Sign survives all
h	H1_female_share	300	299	73.0%	0.0%	4.13	no
hp	H1_female_share	300	300	93.3%	0.0%	5.21	yes
hpn	H1_female_share	300	300	90.7%	0.0%	4.90	yes
h	H2_female_share_in_AUTHORITY	1,164	1,162	67.7%	0.0%	6.64	no
hp	H2_female_share_in_AUTHORITY	1,164	1,162	90.3%	0.0%	4.76	no
hpn	H2_female_share_in_AUTHORITY	588	586	89.8%	0.0%	5.79	no
h	female_share_in_CIVIC	576	576	61.1%	0.0%	6.29	yes
hp	female_share_in_CIVIC	576	576	71.4%	0.0%	7.40	yes
hpn	female_share_in_CIVIC	576	576	70.8%	0.0%	7.20	yes

By entity resolution

Resolution	Estimand	Cells	Positive cells	Sig. positive	Sig. negative	Median slope	Sign survives all
mixed_unknown	H1_female_share	432	432	86.6%	0.0%	4.42	yes
v2_merge	H1_female_share	432	432	86.6%	0.0%	4.39	yes
mixed_unknown	H2_female_share_in_AUTHORITY	864	864	84.0%	0.0%	6.13	yes
v2_merge	H2_female_share_in_AUTHORITY	864	864	81.7%	0.0%	5.97	yes
mixed_unknown	female_share_in_CIVIC	864	864	68.2%	0.0%	6.84	yes
v2_merge	female_share_in_CIVIC	864	864	67.4%	0.0%	6.74	yes

By newspaper handling

Papers	Estimand	Cells	Positive cells	Sig. positive	Sig. negative	Median slope	Sign survives all
all	H1_female_share	300	299	57.3%	0.0%	2.62	no
capped	H1_female_share	300	300	99.7%	0.0%	4.90	yes
excl_evening_star	H1_female_share	300	300	100.0%	0.0%	5.19	yes
all	H2_female_share_in_AUTHORITY	972	966	44.0%	0.0%	2.64	no
capped	H2_female_share_in_AUTHORITY	972	972	99.6%	0.0%	6.04	yes
excl_evening_star	H2_female_share_in_AUTHORITY	972	972	99.9%	0.0%	6.73	yes
all	female_share_in_CIVIC	576	576	8.9%	0.0%	3.80	yes
capped	female_share_in_CIVIC	576	576	100.0%	0.0%	7.26	yes
excl_evening_star	female_share_in_CIVIC	576	576	94.4%	0.0%	7.30	yes

By method

Method	Estimand	Cells	Positive cells	Sig. positive	Sig. negative	Median slope	Sign survives all
-	H1_female_share	900	899	85.7%	0.0%	4.39	no
A	H2_female_share_in_AUTHORITY	864	864	79.6%	0.0%	6.15	yes
A(app)	H2_female_share_in_AUTHORITY	144	144	77.8%	0.0%	6.53	yes
B	H2_female_share_in_AUTHORITY	900	894	84.8%	0.0%	5.87	no
B(app)	H2_female_share_in_AUTHORITY	144	144	83.3%	0.0%	6.27	yes
C@0.3	H2_female_share_in_AUTHORITY	144	144	72.2%	0.0%	3.81	yes
C@0.5	H2_female_share_in_AUTHORITY	144	144	72.2%	0.0%	3.64	yes
C@0.7	H2_female_share_in_AUTHORITY	144	144	77.8%	0.0%	3.66	yes
D@0.3	H2_female_share_in_AUTHORITY	144	144	84.7%	0.0%	5.21	yes
D@0.5	H2_female_share_in_AUTHORITY	144	144	84.7%	0.0%	4.93	yes
D@0.7	H2_female_share_in_AUTHORITY	144	144	83.3%	0.0%	4.60	yes
A	female_share_in_CIVIC	864	864	71.1%	0.0%	6.93	yes
B	female_share_in_CIVIC	864	864	64.5%	0.0%	6.36	yes

Raw vs fixed-effects adjusted

Adjust	Estimand	Cells	Positive cells	Sig. positive	Sig. negative	Median slope	Sign survives all
none	H1_female_share	18	18	66.7%	0.0%	4.64	yes
raw	H1_female_share	864	864	86.6%	0.0%	4.41	yes
topic+paper	H1_female_share	18	17	61.1%	0.0%	1.54	no
none	H2_female_share_in_AUTHORITY	18	18	66.7%	0.0%	5.96	yes
raw	H2_female_share_in_AUTHORITY	2,880	2,880	81.5%	0.0%	5.85	yes
topic+paper	H2_female_share_in_AUTHORITY	18	12	38.9%	0.0%	1.64	no
raw	female_share_in_CIVIC	1,728	1,728	67.8%	0.0%	6.78	yes

Checks added after review

Requested in the mock reviews (research/reviews.md). Wild-cluster bootstrap over the 22 year clusters; logit average marginal effects against the linear probability model; Benjamini–Hochberg control across the 22 role-trend tests, of which 15 survive at q < .05 (against 17 at an uncorrected p < .05).

Estimand	Adjustment	Slope (pp/decade)	Wild-cluster p	Two-way analytic p	Year clusters
H1_female_share	none	+3.11	0.000	0.118	22
H1_female_share	topic+paper	+0.47	0.358	0.595	22
H2_female_share_in_AUTHORITY_B	none	+3.08	0.001	0.107	22
H2_female_share_in_AUTHORITY_B	topic+paper	−0.10	0.875	0.914	22

Estimand	Adjustment	Logit AME (pp/decade)	LPM slope (pp/decade)
H1_female_share	none	+3.08	+3.11
H1_female_share	topic	+1.84	+1.86
H2_female_share_in_AUTHORITY_B	none	+3.13	+3.08
H2_female_share_in_AUTHORITY_B	topic	+2.62	+2.53