Table 2 Coefficient estimates—First expert panel evaluation.

From: Gender diversity of research consortia contributes to funding decisions in a multi-stage grant peer-review process

 

(1)

(2)

(3)

Female ratio

−0.197**

−0.179**

−0.213***

 

(0.079)

(0.076)

(0.081)

Gender balanced

0.011

0.001

0.013

 

(0.058)

(0.062)

(0.061)

Productivity

0.040

0.040

0.022

 

(0.030)

(0.026)

(0.028)

Star scientist

0.113*

0.056

0.094*

 

(0.058)

(0.055)

(0.053)

Cognitive proximity

0.118**

0.064

0.135**

 

(0.046)

(0.043)

(0.054)

Research diversity

−0.002

−0.092*

−0.123***

 

(0.048)

(0.052)

(0.039)

Network proximity

−0.026

−0.067

−0.211**

 

(0.101)

(0.098)

(0.084)

Size

 

0.124***

0.079*

  

(0.040)

(0.047)

Age

 

−0.163

−0.208

  

(0.173)

(0.172)

Shanghai ranking

 

0.080**

0.097***

  

(0.033)

(0.032)

Private partnership

 

−0.085

−0.090

  

(0.072)

(0.071)

Past EUROCORES grant

 

0.127**

0.154***

  

(0.052)

(0.050)

Budget requested

 

0.053**

0.027

  

(0.023)

(0.021)

Number of countries

 

0.156***

0.115**

  

(0.037)

(0.045)

Female ratio panel

  

0.023

   

(0.198)

Productivity panel

  

0.056

   

(0.056)

Panel size

  

−0.154***

   

(0.059)

Age panel

  

0.339

   

(0.389)

Panel workload

  

−0.216***

   

(0.049)

Shanghai ranking panel

  

−0.030

   

(0.186)

Programme fixed effects

 

Year dummies

Domain dummies

Observations

1347

1347

1347

AIC

1532.531

1468.846

1511.633

BIC

1777.196

1749.950

1673.008

Log likelihood

−719.265

−680.423

−724.817

Deviance

1438.531

1360.846

1449.633

  1. This table reports marginal probability effects of the probit estimation, while holding all regressors at their mean value. The dependent variable is ‘Passed 1st Stage’ as defined in Table S1. All other explanatory variables are also defined in Table S1. The three models include different configurations of explanatory variables to test the sensitivity of the estimation. Model (3) is used as the first step in Heckman’s selection model. Robust standard errors in parentheses, clustered at programme-level: ***, **, *, indicate significance at the 1%, 5% and 10% level, respectively.