Last year, 2021, was a shake-up for US labor markets (to say the least). One of the most prominent metrics from the BLS was the number of people quitting and switching jobs (i.e. the Great Resignation). One way to think about it is that in September 2021, ~25% more workers quit their job than in September 2019. [Sidebar = I also love when people counter the mass media headlines and put stuff into context saying the quit numbers are blown out of proportion]. Quitting is part of the shake-up, but what’s more interesting to me is adoption of more remote work. As part of that shake-up, many workers realized they really dislike commuting to an office every day. One survey indicates that more than 50% of “professionals” will look for a new job if their company doesn’t offer a remote work option. This creates a really interesting question for the future of work: will the US population redistribute themselves if job proximity is no longer a constraint?
In this article about identifying the best towns to move to in Connecticut, I asked readers if they intended to relocate in 2022. As of January 15th, nearly 50% of respondents said either “Yes” or “Maybe”. My initial reaction was “wow, that’s a lot!” Upon reflection, I thought “a lot compared to what?”
The US Census Bureau reported data that showed nearly 9.8% of Americans moved in 2019 (prior to the pandemic). It’s not a great comparable apples-to-apples base rate for the survey results, but it is a starting base rate. Two glaring issues. First, the readers of this blog (survey participants) are probably biased toward work that relatively easily allows them to relocate and, second, my survey is small (only 39 responses). So, we need some adjustments to put the poll results in context.
That Census Bureau data is broken down by age, education, marital status, etc. Looking deeper into that data, some groups have higher relo rates than others. It’s possible that the group of people that responded to my survey have a historical relo rate in the 15-20% range. So, I’ll say that I wouldn’t be surprised by a number around 17.5%.
Having a base rate I’m comfortable with, I need to address the uncertainty associated with plans. People say they are going to do lots of things that never materialize for one reason or another. It is sort of a shot in the dark, but I’ll guess that of the people that replied “Yes”, 2/3rds of them will actually relocate and of the people that replied “Maybe”, 1/3rd of them will relocate. That brings the ratio of relocators from 49% to 26%.
Given our base rate of 17.5%, if I did the survey again in a different forum and received 39 responses, how likely would it be that 50%+ of respondents would say they intend to relocate in 2022? While long winded, it’s important to determine this in order to know how surprised I think I should be by the poll results (especially if I am using them to make decisions).
Creating a toy simulation (documented here with lots of notes) with 1,000 instances, it turns out that I should expect to see a number of 50%+… never. It’s zero. The probability is 0.00000000001% (<– I’m just trying to represent an insanely small value with this). Visually, in the heatmap below that represents all 1,000 instances, if an instance of the simulation returned a value of 50%+, the box would be blue. No blue boxes. For me, this is the essence of statistically significant – given data for what we’ve seen historically & even with this small sample size, it is infinitesimally unlikely that I would see these poll results by chance.
# What If 🤔
My initial, gut reaction of “wow, that’s a lot!” appears vindicated. But, what could change that would lessen that sentiment? Many things, but let me point out two.
First, what if the historical relocation rate is actually higher for this demographic of people. Instead of 17.5%, what if it was more like 30%? Running the simulation again with that new base rate would show me that I could expect to get that response metric of 50%+ about 0.4% of the time. That’s tiny and should reinforce the fact that these poll results had “almost no chance” of occuring. Visually, we now have some blue boxes…
Second, what if people were a little more precise on their estimates of the future. Instead of a number of 50%, what if we saw a metric of 26%. Running the simulation again with the original base rate would show me that I could expect to get that response metric of 26%+ about 7.0% of the time. That’s still super small and should reinforce the fact that these poll results were “highly unlikely”. Visually, we now have a few more blue boxes…
# What We Can Do with Small Data
It’s not a great idea to make predictions based on one data point. However, I am happy to say that I think this one data point is a relatively strong one. Other polls and surveys being taken around labor market migrations indicate a similar attitude around remote work – many people want to be free (from a daily office routine) if they can be. Naturally, this creates the prospect of a geographic redistribution of the US population.