The dataset used had 10,174 rows of self-reported gender pay gap data from gov.uk, and was accessed using PosgreSQL. I started by filtering for unique values in the table to see how many individual employers there were (10,174), and then I analyzed which features in the data table had too much missing data to consider further in the analysis. One of the greatest discrepancies here was whether or not 0 values in certain columns truly were null values or not. On the gov.uk website, employer calculations for pay bias towards women or towards men resulted positive or negative percentages with two decimal places. To me, this meant that columns related to such calculations would represent neutral or no bias as 0.0, so I considered this as a value and the integer value 0 as null.

Next, I started querying the dataset to get an understanding of the nation’s pay gap overall. I decided that the best way to comprehend any gaps as accurately as possible was to look at differences in mean hourly percentages so that I could see common salaries and not have my data skewed by outliers on either end of the pay spectrum. I used a CTE to remove null values in the table, and then calculated the average of the median bias percentage across all listed companies to find average 12.82% rate. This meant that on average, men in the UK were reported to earn 12.82% more than women across all companies.

Though this number was useful, it was too large a generalization to use individually and didn’t take into consideration the relative presence of men or women in particular industries (though in turn the argument could be made on the devaluation of wages and salaries in industries with higher female employee rates). The next step was to look at which companies in the country had the biggest pay gaps. I filtered the table for employer names, SIC codes (industry specifying codes), and median bias percentage, and ordered the table in descending order of this percentage. I limited the results to only 10 rows to look at the top 10 companies with the largest gaps, and found them to be mostly small construction companies with an overwhelming majority male employee base. I ran a similar query to this one, adding WHEN-THEN conditions to display whether or not pay differences were skewed to towards either side within pay quartiles. Aside from the almost exclusively male companies, there were some that had equal numbers of male and female employees in each quartile while still showing biases, implying discrepancies somewhere in median earnings.

The next phase in my querying involved looking at bias rates across different geographical regions of the UK. I started by comparing the pay gaps in and out of London by taking a substring of the first two characters in each employer’s postcode. I checked if the substring matched a list of values corresponding to the different postcodes in London (E1, WC, N9, SW, etc.) and found the average pay-gap in and out of London to be 14.20%. and 12.45% respectively. I added an extra query that looked at London in and out of its financial center (City Square Mile) and found a much higher gap rate of 17.42%. I ran similar queries for different cities around the country like Birmingham and Manchester by matching the postcode substrings to the respective city’s first two postcode characters, and found these values to be more similar to the out-of-London averages.

Lastly, I looked at how these gaps might differ by industry. I was most curious about the financial sector and education, as these industries are heavily dominated by men and women respectively. Once I knew their corresponding SIC codes, I filtered the table for employers with these codes and took their average bias percentage as well as a the relative biases within each pay quartile. Both industries had incredibly high pay gaps, with 24.98% for education and 30.82% for finance.