What is the problem this is trying to solve?

The Inter Parliamentary Union release a report each year detailing changes in the representation of women across the world. In 2017 women represented 23.4% of all MPs. Their 2017 report can be read here.

A key problem of international comparisons of the representation of women is, as Miki Caul points out, it "overlooks the fact that individual parties vary greatly in the proportion of women MPs within each nation". Similarly, Lena Wängnerud argues "cross-country studies tend to miss variations between parties within a single system. Variations in the proportion of women to men are even greater across parties than across nations".

This minisite tries to explore what new information exploring party information alongside gender tells us about the representation of women in different countries.

How does it work?

To explore how parties effect gender representation it’s not enough to look at the gender ratios of all the parties individually, as those with the best proportional representation of women are often quite small - for instance, the Green Party in the UK has 100% female representation, in the form of its one MP.

This problem requires a method to determine which parties are the ones to focus on, and it needs to work in very different party systems.

Our approach is to look at the respective contributions to the total gender ratio. For each party, the analysis calculates how much better or worse the proportional representation of women would be if you removed that parties MPs.

After this step, the party that has the highest positive effect on the gender ratio is the Best Contributor (BC) , and the one with the highest negative effect on the gender ratio is the Worst Contributor (WC) . This usually has the effect of identifying the two major parties in a country, but allows flexibility across time and geography.

Knowing the 'Gender Ratio Minus the BC' and the 'Gender Ratio Minus the WC' gives a range of party representation around the national figure. Where this range is large, it shows that there is a big divide inside a country; where the range is small, the parties are relatively aligned on their gender ratio. This can be explored here.

These ratios can also be understood describing what the proportion of women would be if these adjusted their proportion to the average of all other parties - and this is how it is described on the comparisons between countries.

We also calculate for each legislature what the overall proportion of women would be if all parties adjusted their ratio to match the MVP (except where the party already had a higher ratio). This gives us two potential measurements of improvement - how much just the worst performing party improving to the average improves the total gender ratio (Worst To Average), and how much all parties matching the best contributor affects the gender ratio (All to Best).

Why are there fractional people?

These statistics are over the 'legislative period' (parliamentary term, time between elections, etc) - in this time a woman might resign and be replaced by a man (or vice versa).

To capture this the 'value' of a politician is the amount of time they were in the legislature divided by the total length of the legislative period. A women who represented for an entire term would increase the 'women' column by 1 - whereas a man who resigned halfway through would only increase the 'men' column by 0.5.

As there are often seats left vacant for at least some time, the 'total' is often fractionally less than the total number of seats.

Where does the data come from?

The data is sourced from mySociety's EveryPolitican.org - this in turn sources data from a variety of national sources and other data stores like Wikidata.

For this experiment, countries imported from EveryPolitician were limited to those with more than two parties, and with legislatures with 50+ members.

Local council data for the UK is imported from opencouncildata.co.uk. Gender was then derived from name for around 90% (which will have individual inaccuracies but broadly has sufficient usefulness) - the remaining 10% were coded manually based on information on council websites.

Handling Gender Information of Unknown Source

Gender information for EveryPolitician comes from a number of different sources. In some cases it will be directly sourced from the legislative body in question, in other cases information from Wikidata is used. For some individuals gender information will have been gathered through mySociety's Gender Balance game - which requires five votes and 80% agreement to associate a gender to an individual.

Given this mix, it's worthwhile considering the appropriate use of this data. While likely to be more accurate than gender information derived from name (which mySociety use in other research projects), similar considerations apply in that while it would be problematic to take individual action based on gender information of unknown providence, in aggregate errors can be accepted as understanding of the collective without specific action related to individuals.

This is part of the reasoning for restricting legislatures to those with more than 50 members - as this will reduce the risk of errors in gender identification changing the result in a way that has a significant effect on the analysis.

Representation of Transgender Politicians + Intersection with other forms of under-representation

Another issue in analysis of gender information is handling the expression of other gender identities than male and female. For instance, while a politician may choose not to make their gender transition a component of their political identity, there are a few known past and present transgender politicians at the national level – and two of these are included in countries and time periods covered by the EveryPolitician dataset (Anna Grodzka, Poland and Georgina Beyer, NZ).

While Wikidata allows the expression of more full descriptions of gender identity than 'male', 'female' (some discussion of the approach here) - this is reduced when imported into EveryPolitician and both instances of 'transgender female' are changed to 'female'. This conceals in the presented data the intersecting but distinct problem of representation of trans-individuals at the national political level. While such representation is currently insufficient for cross-national comparisons, it merits a discussion as the information is being discarded as part of the analysis process.

This of course isn't a unique issue to transgender representation, looking at the representation as women as a single figure obscures the important role of social factors as such class or race in shaping which women are represented. Creating a metric for comparison across many different countries is inherently reductive and discards important information about local context in every instance.

That said, the aggregate figure remains useful for cross-comparison and question forming within individual contexts. This analysis has attempted to re-complicate the international comparison by moving away from a single national statistic for representation in a way that assigns agency to political actors within each country. Variations among these actors (and international variations in this variation) shows that representation of currently under-represented groups isn’t a natural fact of life in a given country, but reflects choices made – and that other choices can lead to different outcomes.

Known issues:

The dataset is not comprehensive

EveryPolitican does not contain gender information for every country, and not every country has a party system that makes this analysis useful. It is currently limited to 90 countries with sufficient data availability to validate the approach. Countries like Rwanda with high representation of women are not included because they don't have gender information inside EveryPolitican.

Different years are being compared

Currently this compares the latest avaliable term for all available legislatures - and so may not be comparing in exactly the same time range. There may also be changes in the real world that have not yet been reflected in the data source.


If you have any comments on this minisite or the method above, please leave a comment on the feedback page or get in touch at research (at) mysociety dot org

mySociety research newsletter

Sign up to our mailing list for updates about new publications and events.