A data-science study of how visible women's economic activity is: measurement model, uncertainty, sensitivity, scenarios and value of information - Turkiye

This paper, the seventeenth of nineteen in the G20 member series, measures how visible women’s economic  activity is in Turkiye using the Gender Economic Visibility Index (GEVI). The index converts 13 inputs into  parity scores, averages them within five dimensions, and combines the dimensions by a weighted geometric  mean, so that a strong dimension cannot fully offset a weak one. Each input carries an evidence grade that  sets its uncertainty, and the paper carries that uncertainty, together with uncertainty about the weights,  through 10,000 joint Monte Carlo draws. It then tests the result against seven weighting schemes and ten  alternative specifications, attributes the gap to full parity with Shapley values, and runs eight scenarios, a  greedy improvement pathway and a value-of-information analysis for the data. 
Turkiye scores 58.2 out of 100, in the Partial band, with a 90% interval of 54.4 to 61.5 and a rank of 17 of 19  that does not move across the draws. Its strongest dimension is job quality and seniority (0.799) and its  weakest is institutional visibility (0.349), which accounts for the largest share of the 41.8-point gap to full  parity: 12.2 points, or 29%. All 13 inputs are measured (grade A) and none is imputed, so weight uncertainty,  which explains 77.2% of the output variance, drives the spread rather than imputation. Across the  alternative specifications the rank stays at 17 and the score changes by at most 3.8 points. Of the six single lever scenarios, halving the gap to parity on labour access gives the largest gain, +5.3 points, and a greedy  pathway reaches the 65-point threshold after eight steps. The priorities that follow are largely about data:  publishing ministerial and board appointments by sex, recording owner sex on business registrations, and  adding a sex field to wage-tax and taxpayer records, since no sex-disaggregated taxpayer statistic was found  and the index does not yet measure visibility in the tax system directly. 
The weights are a prior and the grade-based standard deviations are judgemental settings, not estimates.  The scenarios show what the index would do if an indicator changed, not that a policy would change the  indicator. Scores and intervals are research estimates, not official statistics.