Covid-19 is a reminder to beware of seasonality in data

A season for change.
A season for change.
Image: Reuters/Danish Ismail
We may earn a commission from links on this page.

From immigration patterns to household spending to music listening, coronavirus is turning 2020 upside down.

In the US, most of that change occurred between February and April, which makes looking at trends tricky. Many data points change from February to April every year, so to understand the magnitude of a change it is necessary to compare it to what happened in previous years. In other words: Beware of seasonality.

Image for article titled Covid-19 is a reminder to beware of seasonality in data

Take marriage trends. Coronavirus has certainly impacted the number of people tying the knot. Many people have postponed getting married so they could have non-socially-distanced ceremonies with family and friends. The US doesn’t report monthly data on marriages, but the state of Florida does. That data show the number of marriages dropping from 14,759 in February to 4,231 in April. (Sociologist Phillip Cohen was the first to point out this trend.)

It’s a huge drop, but to understand just how huge we need to compare the data to previous years. The chart below shows 2020’s trend along with the same trend in the past several years. By comparing the data to the past, we can see that the drop was actually more dramatic than it first appears. Usually, more people get married in March and April than in February.

To account for seasonality, the best way to measure how much marriage declined in 2020 is by comparing the data to 2019. The chart below shows the year-on-year percentage difference for each month. You can see that there was actually a jump in marriages in February of this year compared to last year (perhaps Floridians sensed what was to come). The next few months saw dramatically fewer marriages than last year, with marriages in April almost 70% lower than in 2019.

One note of caution: Looking at year-on-year statistics is not such a great idea if the previous year was an outlier. Had Florida experienced a major natural disaster in 2019 that impacted marriage rates, that year wouldn’t be the best point of comparison. In this case though, we don’t need to worry. Our second chart makes clear that 2019 was similar to previous years, so contrasting just 2019 and 2020 is valid. If 2019 did happen to be an outlier, one way to address that would be to compare 2020’s numbers to the average of the last three years (Phillip Cohen does that in his analysis).

Remember, the abnormalities of 2020 mean that trend analysis next year will also be tricky. Comparing 2021 trends to 2020 generally won’t make sense. Here’s hoping 2021 will be a bit more like 2019.

The following video shows how to use spreadsheet formulas to account for seasonality:

See on Loom