So this graph appears to be from an organization called Marriage Pact, which… look, I don’t care enough to actually research this, but they seem to be AI/economics bros (who also worked at a self-driving car company) who have weird ideas about online dating. Their thing is a “viral dating experience” powered by machine learning.
Their logo is literally on the graph.
As for the study itself, it’s not actually asking how most people meet these days; it’s comparing traditional and nontraditional relationships on a bunch of different axes, to see how people stay together.
Oh, and speaking of axes? You notice anything about that last number? 2020 is what statisticians call a confounding factor. No one was meeting people at the bar, or going out with their friends, or even spending much time with non-immediate family in 2020, even though they very much are in 2024. 2020 was so goddamn weird that it warped the graph for everyone.
How much did it warp the graph? Well, we don’t have the data points…. because the lovely AI bros of the Marriage Pact smoothed out the graph. So we don’t know. But this graph is trying to sell you something.
This is your frienbly remembly to fact check your tumblr posts before you succumb to doomerism. Thenk.
Also, my immediate thought on looking at this graph is how did they do their sampling?
Because there are three options - the first is a longitudinal study or metadata analysis of asking people how they met their partner over the course of 70 years.
It is obviously not this, because the shape is wrong - the curves are too steep. If this is what you’re measuring, a couple who met through friends in 1960 will *still* have met through friends in 1980, and 2000, etc.
But in this graph, around 35% of the couples met through friends from the period of 1980 to 2010, but only about 8% in 2020 - which suggests that 87% of people who met their partner through friends had a break up/died between 2010 and 2020.
This is clearly not the case. So, they are measuring both new relationships started AND how they met, which they could either do:
a) through a longitudinal/metadata study, covering 70 years - which would be a valid approach. Every 2-5 years, ask a sample of new/newish couples how they got together and how long they’ve been together, or compile a bunch of studies over done on this over the years.
This is a sound method and might produce that graph, although it may also explain that sharp online uptick through sampling bias - where are you most likely to advertise a survey, and who is most likely to fill it in.
But… looking at the very small text under the image, I don’t think that’s where they got the sample. They talk about “analysis of the original survey” - singular. Which, unless this is a longitudinal study of 70 years standing, means this is a recent, self selecting sample for a self-report survey, part of which asks people to state the start year of their current relationship, and where they met.
Which means they’ve got a broad sample of people - some of whom have had relationships of up to 70 years in length, and some of whom have clearly met last week (assuming the survey was done in 2020) how they met.
Which means we are counting data of couples who have had their Diamond Wedding as being on a par those of people who have been together six months. Can we say survivorship bias?
Even if they’ve factored in previous relationships (as in “were you dating in this time period, if so, how did you tend to meet people?”) we are veering in to really sketchy self report and there is NO WAY this data can be seen as representative.
For one thing, what is the sample size of people who were dating in 1950 compared to that of 1980? These people must be 85 at least.
And if it is continuous measurement, how have they accounted for the fact that someone dating as a teenager in 1960 would be a 20somethings in 1970 and a 30something in 1980 - which is going to skew their dating and socialising habits ENORMOUSLY?
