How Long Do People Stay at Their Jobs?
The conventional wisdom is that over the past few decades, people are moving jobs more frequently. I figured I could dig in further and try to prove or disprove that using an anonymized random sample of ZipRecruiter’s resumes. Our sample is likely to be non-representative of the U.S. population as a whole, but it might have some interesting insights nonetheless. Given that we’re a fairly young company, we’re only guaranteed to have seen the resumes of people who search for jobs at a cadence of less than once every 6 years or so. Since I don’t have access to the full resume corpus of the US, I’d be making a leap to apply these insights to the labor force as a whole.
Average Tenure
First, let’s look at average tenure per person to get a sense of the distribution. If a person has held 3 jobs for 2, 4, and 4 years respectively, I’ll classify that person’s average tenure as 3.33. Here’s a distribution using that methodology to bucket up resumes by their average tenure:
So the vast majority of people have an overall average tenure of under 5 years, and the preponderance seems to be around 1–2 years. This doesn’t tell us anything about trends over time, however.
An interesting variable is that tenure can be affected by when you held the job in your career. Maybe your first job is always shorter than your fourth job. This plot of Years Since Career Start vs. Year Tenure shows the relative size of each population at each marker.
This one is slightly harder to read, but resembles a violin plot. Generally speaking, the way to read this chart is to go within X-axis category, and compare vertically to get a sense for the distribution within that category. Then, go over to another category to compare distribution versus distribution.
As an example, zooming in on the 2nd year of people’s career, you could view that same vertical distribution on the chart above as this histogram:
And compare it to the 20th year of people’s career, which looks like this histogram:
So there are subtle changes in these distributions based on when in your career you started the job, but by and large, these are all heavily left-weighted distributions that imply most people don’t stay very long, under 3 years or so.
Median Tenure Over a Career
Since these are not normally distributed ranges, I’ll switch to using medians so that we can try to compare each bucket. So instead of the granular “distribution per X-axis category”, we’ll have one data point to compare, the median tenure split by how many years into a person’s career they started the job. Because so many of the data points are below 2 years, I’ve switched to using monthly tenure instead of yearly for a little more granularity as well.
Now that’s pretty interesting! People’s first jobs are actually longer than later ones. I have a hunch that this may be that bias I stated previously, and that people can self-select their first “solid” or “relevant” job back in history, and leave off their “true” first jobs as they’re not particularly relevant or flattering. I know I leave off my first job as a Cashier from my resume when I’m applying to Analytics jobs these days.
Now that we’ve got a single data point to compare, let’s try to break it down based on the year the person started their career. I’ll define that as the start date of the first job I find on their resume. The aforementioned first job bias can also affect this data point, but assuming each age cohort has the same bias, we should still be able to read relative differences.
Each year bucket represents the ending of the 5 years before it. So the 2015 series has data from 2010–2014, etc. It’s also worth noting that because the series is set in 5 year increments, but the X-axis is in yearly increments, each data point underlying these lines represents a declining number of samples. For example, in the (2015,0) data point, we have 5 years worth of data to pull from (2010–2014). However, in the (2015,3) data point, we only have 2 years of data to pull from, because only people who started their careers in 2010 and 2011 can actually have a 3 years into their career at this point. This is true for all these buckets, and we’d need to project how long we think people will stay at those jobs to have an apples to apples comparison to series in which we know how long they stayed.
Since we see that clear deviation from the trend in year 0, let’s dive into that for a bit. Here’s what that difference looks like in bar form:
I’ve already filtered for noise in a lot of this data, so even though we have fewer data points on people who started their careers in the 70s-90s, I’m fairly confident we can say that the median worker who started a few decades ago stayed longer at their first job than workers who started in the 2000s. How much longer? Tough to say because of the first job selection bias, but the rough order of magnitude might be about 2x.
How about another selection from 5 years into their career? Does that show the same trend?
Pretty much! The difference narrows a bit here, the difference between 2000s and 80s-90s is more like 1.5x than 2x.
Just to be clear, this data doesn’t conclusively show that someone who’s actual first job was around 1985 held that job for ~7 years. It says that for resumes who say their first job was around 1985, those jobs lasted 7 years. Hopefully there’s a similar amount of bias in the 2000 data set, with people that actually started work in the late 90s but claim their first job was in the 2000s. The relative comparison between those data points shows that people might be holding jobs for shorter amounts of time near the start of their career. Or who knows, it could be that more recent workers simply have different resume habits, and actually include their first few jobs, even if they’re not quite as relevant.
In general, I think this analysis bears out the conventional wisdom, at least for ZipRecruiter’s sample. Trying to control for career length and start times, we can see a relatively clear pattern in resumes claiming jobs of shorter and shorter lengths. The simplest probably reason why that’s the case? People are holding jobs for shorter and shorter time-frames. As to whether or not that’s a good thing for employers and employees, and why it’s happening, that’s a challenge for another time.
