Table 1 Regression models to predict the gender diversity-association effect in mentorship and co-authorship networks

From: Gender and racial diversity socialization in science

Model

a1

a2

b1

b2

Network

Mentorship

Mentorship

Co-authorship

Co-authorship

Regression type

Linear

Logistic

Linear

Logistic

Diversity using null model

No

Yes

No

Yes

(Intercept)

−0.181***

−3.835***

−0.323***

−6.082***

 

(0.031)

(0.226)

(0.003)

(0.031)

Institutional prestige

−0.037*

−0.225

−0.005***

0.004

 

(0.018)

(0.121)

(0.001)

(0.014)

Advisor group size

0.001

0.004*

0.000***

0.005***

(Number of early junior co-authors)

(0.000)

(0.002)

(0.000)

(0.001)

Researcher is woman

0.083***

0.565***

0.091***

0.710***

 

(0.009)

(0.050)

(0.001)

(0.007)

Women % by subfield

0.830***

5.002***

0.806***

6.826***

 

(0.024)

(0.171)

(0.003)

(0.032)

Women % by country

0.637***

4.779***

0.754***

6.226***

 

(0.070)

(0.525)

(0.005)

(0.051)

Early gender diversity

0.118***

0.314***

0.086***

0.402***

 

(0.012)

(0.046)

(0.001)

(0.006)

  1. Under linear regression (models a1 and b1), the dependent variable is the proportion of women among advisees/junior co-authors of individual researchers in the established period. Under logistic regression (models a2 and b2), the dependent variable is a binary coding of whether individual researchers in the established period have a high percentage of women advisees/junior co-authors compared with the null model. The key variable is the early gender diversity, which measures the proportion of women among advisor’s group members/junior co-authors in the early training period. It uses the raw women percentage in models a1 and b1, while it is a binary variable relative to the null model in models a2 and b2. Two-sided t-tests are used for multiple comparisons. Robust standard errors are given in parentheses. ***P < 0.001; *P < 0.05.
Back to article page