Dealing with skewed mean

Question

I have a data set of districts, farmland area and fertilizer subsidies issued for those areas. I.e, using made up numbers,

district | area | subsidy | subsidy per area (computed)
abc      |   20 |   500   |         25
cde      |   30 |   750   |         25
fgh      | 0.02 |    15   |        750    <--- looks off

I'm trying to visualise the subsidy per area but in districts that have very small amounts of farming the subsidy per area seems abnormal. The nationwide average is pretty much around 25. So, I can safely say that the subsidy amount is directly related to the area being subsidised, which is to be expected as fertiliser usage is dependent on the area being farmed. My theory is that the exception on small areas is due to there being a minimum subsidy amount irrespective of the land area.

Are there any techniques to deal with the above scenario when visualising data?

Answer 1

As per business statement it is a exceptional scenario when minimum subsidy will be provided. So for displaying common behavior of data you can drop these exceptional values from table. You can use box-plot to visualize spread of data and then remove else, if you are aware with max range in normal scenario you can drop rows having value more than that.

Answer 2

If districts are visualized in a scatterplot which subsidy is labeled as y-axis and area as x-axis, subsidy per area should be shown as the slope of the scatterplot. If subsidy per area is around the nationwide average of around 25, the slope of the scatterplot should be pretty much around 25.

You theory of exception on small areas can be visualized as an outlier in this scatterplot. If there is a minimum subsidy amount for example at 15, you should see a low bound in the scatterplot that no districts are having subsidy below 15.

In comparison with the slope 1, the slope 25 is pretty steep, and the slope 750 is extremely steep, which the difference of both slopes cannot be highlighted in normal scale. You may need to rescale either the x-axis or y-axis so that the outlier can be identified easier.