Age Gaps: How much does age matter in dating?

[Regina] (0:00 - 0:18)
There was this one dude, I found, who was 54, but he said he would not date a woman over 50. Of course, she's got to be younger, wouldn't even consider someone his age, but then the matchmaker gave him a 61-year-old.

[Kristin]
Ooh, bold.

[Regina]
Very bold.

[Kristin] (0:23 - 0:46)
Welcome to Normal Curves. This is a podcast for anyone who wants to learn about scientific studies and the statistics behind them. It's like a journal club, except we pick topics that are fun, relevant, and sometimes a little spicy.

We evaluate the evidence, and we also give you the tools that you need to evaluate scientific studies on your own. I'm Kristin Sainani. I'm a professor at Stanford University.

[Regina] (0:46 - 0:52)
And I'm Regina Nuzzo. I'm a professor at Gallaudet University and part-time lecturer at Stanford.

[Kristin] (0:52 - 0:58)
We are not medical doctors. We are PhDs, so nothing in this podcast should be construed as medical advice.

[Regina] (0:58 - 1:11)
Also, this podcast is separate from our day jobs at Stanford and Gallaudet University. Kristin, today we're diving into a paper on a dating topic that sparks a lot of heated opinions, age preferences.

[Kristin] (1:11 - 1:18)
Oh, is this the old stereotype? Men want younger women and women want older men?

[Regina] (1:19 - 1:35)
Right. Some people have really strong preferences, like they absolutely want someone younger than them or older than them, or they have a maximum age beyond which they will not even consider dating. And if you suggest they step outside of that comfort zone, they get surprisingly cranky.

[Kristin] (1:37 - 1:44)
So I'm guessing the big question we're looking at today, how much does age actually matter in the dating world?

[Regina] (1:44 - 2:21)
Exactly. Is it true that love has no age, or is it that no, actually, love is not blind and it does have an age? So the claim for today is this, that age does matter for people on date. And of course, Kristin, we are not talking about crazy situations like 26-year-old Anna Nicole Smith and the 89-year-old billionaire.

We are talking about moderate age differences for normal people.

[Kristin]
Okay. Got it.

[Regina]
Yes. Today's paper, by the way, Kristin, is another one from Paul Eastwick and Eli Finkel, who are the amazing podcasters we encountered recently.

[Kristin] (2:21 - 2:42)
You've been reading a lot of their work lately, Regina. This is the second one of their papers that you've covered. The other paper was the Dating Wishlist episode.

And we should point out that Paul and Eli are researchers, love researchers, but they also host the Love Factually podcast, which examines the psychology behind rom-coms. It's a lot of fun.

[Regina] (2:42 - 3:09)
Mm-hmm. It is. And yes, I might be turning into a bit of an academic stalker.

I don't know. You know, it's just that when I was looking at their Dating Wishlist paper for that previous episode, I came across this new 2025 paper by them. And I was just so fascinated because this topic of age and dating comes up constantly in real life.

Kristin, you and I talk about it a lot.

[Kristin] (3:10 - 3:17)
Well, yeah, we have to because, Regina, you tend to date men who I think are too old for you. So I'm constantly giving you a hard time about that.

[Regina] (3:18 - 3:41)
I like them mature. What is wrong with that?

[Kristin]
Well, there's mature, and then there's mature.

[Regina]
Moving on, it was cool data, though, because they analyzed reports from over 4,500 blind dates.

[Kristin]
Oh, that's a lot of blind dates. A lot of awkwardness.

And big sample.

[Regina] (3:41 - 4:01)
Awkwardness. It is a big sample. And the results were really surprising.

So I just, I couldn't resist bringing them out again. Kristin, I know you are curious about the stats topics. Today we will learn how to really properly interpret regression results.

And I think you'll like it.

[Kristin] (4:01 - 4:09)
It's super important, yes. But before we get to the data, I can tell that you want to give us some historical perspective on this topic. What do you have for us?

[Regina] (4:10 - 4:18)
You know me too well, Kristin. All right. Have you ever heard of the half your age plus seven rule?

[Kristin] (4:18 - 4:25)
Oh, yeah. So that's the minimum age you should date, right? And Regina, are we calculating that for ourselves today?

[Regina] (4:25 - 4:39)
I will admit my age. I am not ashamed. I am almost 54.

And my minimum then is half of 54, which is 27, plus 7, which is 34.

[Kristin] (4:39 - 4:44)
Oh, I love it. You should totally date 34-year-olds, Regina. Absolutely.

Go for it.

[Regina] (4:45 - 4:53)
Kristin, that is 20 years younger. 20. I will just say I am not sure that 34 is the ideal age for me.

[Kristin] (4:53 - 4:56)
Oh, I think you'd have a lot of fun with a 34-year-old.

[Regina] (4:58 - 5:14)
No comment on that. I might be blushing. Yeah.

Moving on. Back to the math. We can also use that formula to figure out the oldest guy that is still allowed to date me according to this rule.

[Kristin] (5:14 - 5:32)
Oh, that is a great math problem. I think I'm going to use that in my PhD interviews. It's like algebra on the fly.

Okay. So if the guy's minimum date age is half his age plus seven, then we got to reverse engineer. So we would subtract seven from your age and then double that number.

[Regina] (5:32 - 5:44)
Very nice algebra. Okay, so we take 54 minus seven, which is 47, then double that, which is, are you ready for this? 94 years old.

[Kristin] (5:45 - 5:52)
94? Okay, Regina, I know you like mature men, but 94 seems really, really extreme.

[Regina] (5:53 - 6:03)
I wonder what the definition of a successful date is with a 94-year-old. Is it like he walks to dinner and he stays awake during dinner?

[Kristin] (6:04 - 6:23)
You're reminding me of the Italian centenarians from the Vitamin D episode. They were considered quote unquote healthy, I think, just because they were still breathing. Regina, this is making me think that this formula, maybe it's just for fun and maybe it's not actually that accurate.

Is it evidence-based?

[Regina] (6:24 - 6:53)
Good question. It actually does have a citation.

[Kristin]
Oh, no way. It's from a paper?

[Regina]
Well, not a scientific paper. It's from a marriage advice book from 1901.

Great title, Her Royal Highness Woman and His Majesty Cupid. It's hilarious. I'll put a link in the show notes.

He gave this age rule as something he had heard recently, but it was for how a man should choose a woman to marry.

[Kristin] (6:54 - 7:02)
Of course. Well, it was 1901, and I guess since women couldn't even vote, they probably didn't get a lot of choice in mates either.

[Regina] (7:03 - 7:29)
Exactly. And, in fact, I found a quote that kind of underlines that. Want to hear?

[Kristin]
Of course.

[Regina]
Never marry a woman richer than you or one taller than you or one older than you. Be always gently superior to your wife in fortune, in size, and in age so that in every possible way she may appeal to you for help or protection.

[Kristin] (7:29 - 7:43)
Oh, my goodness. I'm rolling my eyes. This is very male dominant. And he had issues.

[Regina]
I would say yes.

[Kristin]
But, Regina, getting back to that stereotype that men like younger women, is it true statistically? Are there any data on this?

[Regina] (7:44 - 7:53)
There are actually. United Nations says that around the world, men marry women about four years younger on average.

[Kristin] (7:53 - 8:02)
Regina, I'm wondering, though, you mentioned worldwide data, which, of course, in some countries there definitely is an age gap in marriage. What if you look at just the U.S.? Is this still true?

[Regina] (8:03 - 8:18)
Yep. Still true. Looking at mixed gender couples in the U.S., only about one third of them are within two years of each other, so comparable ages. About half have the man at least two years older, but only one in seven has the woman older.

[Kristin] (8:18 - 8:19)
Interesting.

[Regina] (8:19 - 8:30)
Yeah. And data from surveys, online dating, same thing. Men say they prefer women a few years younger.

Women say they prefer men a few years older. Yep.

[Kristin] (8:30 - 8:32)
Yeah. So along the lines of that stereotype, yep.

[Regina] (8:33 - 8:56)
Yeah. Speaking of online dating, I don't know about you, but I have found men can get very defensive about their age preferences. I have seen these occurrences where they lie about their age just to get matched with younger women.

And if the algorithm dares to match them with someone close to their own age, oh my gosh, can you imagine? They call it age discrimination.

[Kristin] (8:57 - 8:59)
How is that discrimination?

[Regina] (9:01 - 9:04)
They're discriminating against older men that want younger women?

[Kristin] (9:05 - 9:08)
I don't think they understand the word discrimination.

[Regina] (9:08 - 9:15)
No, they do not. I want to be like, come on, dude, this is not a civil rights issue. Just like calm down.

[Kristin] (9:15 - 9:19)
Yeah. Okay, Regina. Now let's get back to the paper.

Tell us more about the paper.

[Regina] (9:20 - 9:31)
Right. It was published in the Proceedings of the National Academy of Sciences in 2025. But just to be clear, Kristin, the study was not about that half your age plus seven rule.

That was just me bringing in pop culture.

[Kristin] (9:32 - 9:42)
But Regina, has anyone ever studied that rule? Like with an empirical actual study?

[Regina]
I don't know.

That's a good question.

[Kristin]
That would make a really good PhD thesis.

[Regina] (9:43 - 9:48)
Great PhD thesis. I want to do that. We would definitely cover it on the podcast.

[Kristin] (9:48 - 9:49)
We would cover it. Yes.

[Regina] (9:49 - 10:07)
Yes. Okay. The study did have interesting data, although not that.

It was data on age preferences and outcomes from 4,500 blind dates. These dates came from one service called Tawkify, which is a matchmaking company with actual human matchmakers.

[Kristin] (10:08 - 10:12)
Oh, that's interesting. So a human rather than a computer algorithm. Maybe that's better.

[Regina] (10:13 - 10:43)
Maybe. So here's how it works, because it's a little different. There are two membership levels, and the first is what's called members.

And at the time of the study, they paid an annual fee of about $100. And for that, they got to be included in a pool of possible dates for the second tier of membership, what's called clients. And the clients paid more money, and they were set up on dates with a guaranteed number of potential partners.

[Kristin] (10:43 - 10:51)
Oh, that's interesting. So the members paid $100. What did the clients have to pay to get this personalized matchmaking service?

[Regina] (10:52 - 11:00)
I've seen that packages go from $4,900 to $70,000 for the ultra top tier.

[Kristin] (11:01 - 11:05)
That is a lot of money. I'm wondering now, what do you get for the $70,000?

[Regina] (11:06 - 11:09)
That must be like the white glove service. They do not say.

[Kristin] (11:09 - 11:11)
It's a big secret. Yeah.

[Regina] (11:11 - 11:19)
I think that is very secret. But really, come on, Kristin, can you put a price on true love?

[Kristin] (11:19- 11:28)
For $70,000, yes, I think I can.

But Regina, they must have very successful and wealthy clients then.

[Regina] (11:28 - 11:53)
That would be a good inference from this, wouldn't it? They do take care of everything for all their clients. It was interesting.

They set up the dates. What they do is send you an email with your matches, first name, a few interests, whether they have kids and their approximate age, you know, like late 30s, early 50s, whatever. They even arrange the location and the time.

[Kristin] (11:53 - 12:03)
Oh, very nice. But is this completely blind? Like you don't have their last name, so you can't Google them and don't send you a picture ahead of time?

[Regina] (12:03 - 12:06)
Totally blind. It's like old school, right? They just probably did it before Google.

[Kristin] (12:06 - 12:07)
Literally blinding, yes.

[Regina] (12:08 - 12:12)
Right. The matchmaker knows what you both look like, but not those going on the date.

[Kristin] (12:13 - 12:13)
Interesting.

[Regina] (12:13 - 12:25)
So, then you do your date, and then afterwards, both people fill out a feedback questionnaire so the matchmaker can use the information to keep refining future matches if needed.

[Kristin] (12:25 - 12:31)
Oh, but they have a questionnaire, so is that where the researchers got the data for this study?

[Regina] (12:32 - 12:58)
Exactly. I love that the researchers were able to use real-world data. I think this is great because all the daters had an incentive to be honest on the form.

It's not like you're asking a bunch of hungover college students, you know, hypotheticals in a lab looking at photos. And I read the literature a lot, and this kind of real-life honest data is hard to find in dating research.

[Kristin] (12:58 - 13:07)
But, Regina, how did the researchers get access to the data? Because this is a private, proprietary company. How did Eli and Paul get their hands on the data?

[Regina] (13:08 - 13:35)
Yes, this is interesting. Good question. The senior author on the paper was at the time the CEO of Tawkify.

Oh, interesting. Mm-hmm. In the paper, they talked about this.

They said that the three other authors, Paul and Eli and another researcher, they consulted on the questionnaire design, and they shared their findings with Tawkify, but they did not get paid for it, and that the CEO had no say in the data analysis or the publication.

[Kristin] (13:35 - 13:45)
Oh, interesting. So, we do have some conflicts of interest, but they tried to keep it fair and balanced, and I guess Eli and Paul get a paper out of it, and Tawkify gets some information.

[Regina] (13:46 - 14:07)
Right. You can see how this would be mutually beneficial because for Tawkify, the result would be useful. They are probably internally asking themselves things like, how much should matchmakers keep age in mind when setting up the couples, and should they treat clients' age preferences as a suggestion or a hard rule?

[Kristin] (14:07 - 14:17)
Oh, and this is really cool because actually the company doesn't have a vested interest in one answer over the other. They just want to know what works, and that actually helps reduce bias in this study, Regina.

[Regina] (14:18 - 14:39)
Yeah, it's a clever study design. I love it. Before we get into the details about the study, though, do you want to hear some fun things about Tawkify?

[Kristin]
Of course.

[Regina]
Of course. On their website, they say that you are 6.5 times more likely to find your best match with Tawkify than on a dating app, which seems very precise.

[Kristin] (14:39 - 14:45)
That sounds very made-up, actually. Like, do they have a reference? Did they do a study?

Where is that number coming from?

[Regina] (14:45 - 14:56)
I know. There was no citation. We've got to hunt that one down.

They also said that something like 80% of people on Tawkify are matched within 12 dates.

[Kristin] (14:57 - 15:08)
Right, like after they've spent $200,000. But what's the outcome here? Matched meaning, like, they found the love of their life, they got married, what was the outcome?

[Regina] (15:09 - 15:12)
Conveniently, they did not define matched. Of course.

[Kristin] (15:12 - 15:18)
Regina, though, you were thinking about signing up for this. Did you end up signing up?

[Regina] (15:19 - 15:28)
I was curious, so I filled out the intake form and then had an intro phone call with a matchmaker-slash-salesperson, but I have not signed up.

[Kristin] (15:29 - 15:30)
I want to hear about this.

[Regina] (15:30 - 15:46)
It was interesting. And when the guy asked how I found Tawkify in the first place, I said, I read the CEO's paper in the proceedings of the National Academy of Sciences. Oh, wow.

[Kristin]
What did he say to that?

[Regina]
He thought it was hilarious. I don't think they get many referrals in that way.

[Kristin] (15:47 - 16:13)
Yeah, I doubt they were counting on their academic journal paper as a recruiting tool. But, you know, Regina, we are giving them some free advertising on the podcast here. I think that you should get a discount for the client's membership.

All right, let's get back to the study now. So Eli and Paul have this data from Tawkify on how well blind dates went. How did they use this data to study age and dating?

[Regina] (16:14 - 16:19)
I want to talk about study details, but I'm going to draw out the suspense. I think it's time for a short break.

[Kristin] (16:28 - 16:38)
Regina, I've mentioned before on this podcast our introductory statistics course, Demystifying Data, which is on Stanford Online. I want to give our listeners a little bit more information about that course.

[Regina] (16:38 - 16:49)
It's a self-paced course where we do a lot of really fun case studies. It's for stats novices, but also people who might have had a stats course in the past but want a deeper understanding now.

[Kristin] (16:49 - 17:22)
You can get a Stanford Professional Certificate as well as CME credit. You can find a link to that course on our website, normalcurves.com, and our listeners get a discount. The discount code is normalcurves10. That's all lowercase.

Welcome back to Normal Curves. So Regina, we've got all this great data on how a bunch of blind dates went, and now you're going to tell us how the researchers used this data to study age and dating.

[Regina] (17:22 - 17:46)
Well, the matchmakers keep track of client age preferences, but the matchmakers don't always stick to them. Sometimes they spot a promising match who's older or younger than someone's age range, and they set them up anyway. So that means people end up going on dates across a pretty broad span of ages, much broader probably than if they were choosing for themselves.

[Kristin] (17:47 - 17:53)
Oh, this is kind of cool, Regina. It's basically like a natural experiment because the clients don't get to choose for themselves.

[Regina] (17:54 - 18:01)
And that was not randomized, though. We're not sending someone to a date with a 94-year-old one week and a 34-year-old the next.

[Kristin] (18:01 - 18:13)
That's a good point. But, oh my goodness, that would be a great reality TV show if the dates were totally randomly assigned like this, regardless of age. I would so watch that.

[Regina] (18:14 - 18:19)
So instead of love at first sight, it would be what? Love at first hip fracture?

[Kristin] (18:21 - 19:11)
Okay. Yeah, important point, it's not randomized, so it's not the perfect experimental design. Ideally, we would like things to be totally randomized where the matchmaker just randomly assigned everybody regardless of age.

And I want to point out here, Regina, that because it's not randomized, we can't draw causal conclusions from these data. We may, at times in this episode for ease, use a little bit of causal language, but let's keep in the back of the mind that this is observational. One thing that's nice, though, is we do reduce some biases because the daters don't get to choose their partner's ages.

And it's also really nice that the daters didn't get to choose their partners because we get a window into what happens when people are nudged outside of their comfort zones. But Regina, how far outside are we talking? Were these matchmakers just fudging the age limit by a couple of years or did they go big sometimes?

[Regina] (19:12 - 19:33)
Good question. On average, the age gap was only about three years, although some parents stretched 20 or even 30 years. Yes, I'll talk about that.

But as far as being outside the age limits, only about 15 percent of the dates had at least one person older than their partner's maximum preference.

[Kristin] (19:33 - 19:44)
OK, so that's enough rule breaking to give you some real variation, especially because the sample is large, 4,500 blind dates, you said. Do you want to break down that 4,500 a little bit more for us, Regina?

[Regina] (19:45 - 20:13)
Yep. Like I said, 4,500 dates, so nearly 9,000 reports because they got reports from both sides. Right.

[Kristin]
There's two people on a date.

[Regina]
Right. So twice as much data, yeah.

So that was 6,300 people because some people went on more than one date. And it was roughly half men, half women. The ages of all of these people ranged from 22 to 85.

The average was around 46 and about two-thirds of them were white.

[Kristin] (20:13 - 20:28)
OK, so we've got this big data set with dates across a wide range of ages and sometimes within people's preferences, sometimes outside of people's preferences. And what were the outcomes? Like what was on this questionnaire?

What did they ask them about how well the date went?

[Regina] (20:28 - 20:45)
Yeah, they had three outcomes. The first was what they called romantic attraction, which was the average of three statements. I enjoyed spending time with my date. I was attracted to my date. And they seemed like a great romantic partner, all on a one-to-five scale.

[Kristin] (20:46 - 20:48)
Getting a little bit at sex there.

[Regina] (20:48 - 20:49)
Yeah.

[Kristin] (20:50 - 20:52)
Is that what we're talking about?

[Regina] (20:52 - 21:04)
You interpret that however you want. OK. The second was the overall rating of the date, and it was on a one-to-five scale from awful to wonderful.

[Kristin] (21:04 - 21:09)
OK, so that might include more than sex, like, you know, how the conversation was.

[Regina] (21:10 - 21:17)
The conversation. The third was the simplest one, would you want a second date, yes or no? And if they both said yes, then they got another date.

[Kristin] (21:18 - 21:25)
Oh, I like that measure because it's very straightforward. Like you either wanted a second date or you didn't. And so that's going to be very honest and telling, I think.

[Regina] (21:26 - 21:32)
Here's a fun insider detail, by the way. When both people say yes, the matchmaker actually gets a bonus.

[Kristin] (21:33 - 21:54)
Oh, so they have a financial incentive to do well in the pairing. That's great. The company pays them the bonus, though, not the client.

[Regina]
Not a tip.

[Kristin]
Well, I was wondering about that. So if you're paying $70,000 or some high amount of money, do you have to tip the matchmaker if they do a good job?

What's the etiquette of matchmaking services?

[Regina] (21:55 - 22:36)
I'll tell you. No, I won't. I'm not signing up for the $70,000 package.

OK, Kristin, now let's talk results. There's a lot in the paper, but I'll just walk us through the highlights. They really had just two big questions.

First, does a partner's age in general affect how highly they are rated? Like do women rate older male partners more favorably? And do men rate younger female partners more favorably?

Second question, does it matter if someone's date, their partner, was inside or outside of their stated age range preference?

[Kristin] (22:36 - 22:41)
OK, so those are the big questions they're trying to answer. How did they actually handle the data?

[Regina] (22:41 - 22:54)
Each date generated two observations, like we talked about. One rating from each person. We're going to talk about the person doing the rating and then their partner.

And many people went on multiple dates.

[Kristin] (22:55 - 23:08)
Oh, so Regina, we have multiple sources of correlation here. These are correlated observations, and that's something we've talked about before, back in the pheromones episode. So did they handle the correlated observations correctly, Regina?

[Regina] (23:08 - 23:19)
They did. This is a problem that comes up a lot. And yes, they did it correctly with random effect models, which we also talked about in our dating wishlist episode.

[Kristin] (23:19 - 23:27)
Excellent. So the statistics police are satisfied here. We get really upset if you don't handle your correlated observations well.

[Regina] (23:29 - 23:45)
They do. And here's a cool part. They also pre-registered their analyses. Yay.

And they also shared all of their data and their R codes. So I can rerun their analyses and even do some of my own. We'll see in a few minutes why that matters.

[Kristin] (23:46 - 23:55)
This is great. We love it when researchers share data in code. It makes me trust their results more and also gives us the opportunity to rejigger things if we feel like we need to.

[Regina] (23:55 - 24:19)
Yep. Okay, so let's start with that first big question, the effective age in general. Kristin, this is where I got a little puzzled.

I kind of scratched my head. So they built a linear regression with the partner's actual age as the predictor and then how highly they got rated as the outcome. And the researchers fit a straight line for that separately for men and for women.

[Kristin] (24:19 - 24:32)
Okay, but wait a minute. The only predictor was the absolute age of the partner?

[Regina]
Yep.

[Kristin]
That's weird. Shouldn't we use relative age or the age gap between the partners rather than absolute age? I don't quite get that.

[Regina] (24:32 - 25:04)
Right, exactly. This is what I thought in my notes. I put down, wait, what?

Why? That didn't make sense. So I reran it myself using that relative age difference, the age gap.

And I also did it another way using both the person's age and the partner's age as separate predictors in the model. And you know what? All the results, the patterns were pretty much identical, surprisingly.

Same trends, same statistical significance. However you slice it, like the results hold.

[Kristin] (25:05 - 25:21)
Okay, so that's good because it's robust. And I'm thinking maybe that's because age and age difference are correlated. Like younger people tend to date a narrower range of ages and older people date a wider range.

And since those things are correlated, it doesn't matter what you put in the model, maybe?

[Regina] (25:21 - 25:33)
Exactly, exactly. So here's the big picture for both men and women. The older the partner, the lower the ratings on that romantic attraction and overall quality of the date.

[Kristin] (25:34 - 25:39)
Well, here's the age discrimination then. We're all getting penalized. That's the age penalty.

[Regina] (25:41 - 25:55)
Right. Those guys, they just, they don't get it. Older gets the age penalty.

Exactly. And these, by the way, were highly statistically significant. And they looked to see if there was a difference between men and women.

Nothing. No difference between men and women on this.

[Kristin] (25:56 - 26:04)
Oh, this is super interesting, Regina. So it goes against the stereotype of women liking older men. In fact, women wanted their men young.

[Regina] (26:05 - 26:11)
I'm thinking, Kristin, I might need to admit it. You might be onto something with me and that 34-year-old.

[Kristin] (26:11 - 26:24)
You need to date a 34-year-old, Regina. Absolutely. But Regina, you said statistically significant.

As statisticians, we always have to ask, how big was the effect? How big was this age penalty?

[Regina] (26:25 - 26:48)
Good question. They reported the effect sizes as standardized betas. And maybe we can unpack those in a moment.

But for the magnitude, they were all similar in size. Let me give you one concrete example that we can interpret. For men rating that romantic appeal of their date, the beta was negative 0.07. And Kristin, lead us through what that means. Right.

[Kristin] (26:48 - 27:38)
So first of all, beta coefficient, that is just a fancy word for slope, the slope of the lines that you just talked about. But this is a standardized beta coefficient. And so negative 0.07, that's actually really, really small. In our rough rule of thumb labels that we've talked about before, a standardized beta of 0.1 is considered small. So 0.07 is even smaller than small, basically minuscule. And let's look at the precise definition.

Remember, again, this is a slope and it's in standard deviation units. So we could interpret this as for every one standard deviation increase in a woman's age, the man's romantic rating dropped by 7% of a standard deviation. And Regina, I think people aren't going to really understand what this means because I don't even know what is a standard deviation of age in this sample.

We need to know that.

[Regina] (27:39 - 27:46)
Yeah. Standard deviation of women's ages, by the way, was 11.3 years. Oh, but that's a little arbitrary, right?

[Kristin] (27:46 - 28:22)
The standard deviation here is totally dependent on the sample. Like if we had a sample that was just 40-year-olds, it would have a tiny standard deviation. If we had a sample where the ages spanned, you know, 18 to 85, you're going to have a much bigger standard deviation like we do here.

So it's really hard for me to interpret that standardized beta coefficient out of context. And I'm wondering, why did they bother to standardize here? We have natural units that make sense.

It's like the number of boats we talked about in the dating wishlist episode. It's a nice concrete unit. We have years of age and everybody understands that.

So I don't understand why they would turn this into standard deviation units.

[Regina] (28:22 - 28:33)
Exactly. And that is why I went back and redid the math to get those coefficients, those slopes in plain units. It's a useful trick.

I'll put in the details in the show notes for people.

[Kristin] (28:34 - 28:38)
Oh, fantastic. Yes. In years, something we all understand.

So what did you find?

[Regina] (28:38 - 28:50)
All right. For every extra year a woman is older, her romantic attractiveness score drops by about 0.007 points on a 1 to 5 scale.

[Kristin] (28:50 - 29:00)
0.007 points? That is absolutely minuscule. Although a year of age is not really a ton.

What if we looked at that over a decade, for example?

[Regina] (29:00 - 29:07)
Yep. Yep. A whole 10 years only drops the score by 0.07 points on a 1 to 5 scale.

[Kristin] (29:07 - 29:16)
Which is still basically nothing. And that's a great trick, Regina, by the way, when interpreting beta coefficient slopes. You want to zoom out to a meaningful time span.

It's more interpretable.

[Regina] (29:17 - 29:46)
Here's another meaningful time span. In the study, the average age gap was three years. So three years would be a 0.02 point drop in romantic appeal. Meaning it basically didn't matter. But, Kristin, I decided to push the model just for fun. And I asked, how long would it take for a woman's score to actually fall by a full point, one point, like from a 5 down to a 4?

[Kristin] (29:46 - 29:59)
Oh, this is a little fun math. And we can put some details in the show notes for those who like math. But we have to say you need to be really careful with this exercise.

You are extrapolating, Regina. That is a cardinal statistical sin.

[Regina] (29:59 - 30:14)
It is totally extrapolating. Illegal in stats court. The stats police are going to come arrest me.

But I'm doing it to make a point about how tiny the effect really is. So the answer, how long would it take a woman's score to fall by a full point? 157 years.

[Kristin] (30:16 - 30:24)
So compared to a 20-year-old, a 177-year-old might be docked one point in romantic attractiveness.

[Regina] (30:26 - 30:44)
Isn't that great? I love that. OK.

And then I pushed it even further. What if they started out as like a hottie 5, right? How many years would it take for them to drop to a rock bottom 1?

That would take 628 years, which I'm picturing as like a preserved mummy, right?

[Kristin] (30:46 - 31:33)
Preserved mummy would probably get a 1 on romantic attractiveness, I hope. Yes. OK, we're just being funny here and making a point about how small these effect sizes were.

I just want to point out, of course, the effect sizes here are based on the sample that we have in front of us, where most of the daters were just a few years apart. So it's not actually fair. We can't extrapolate even to a situation where a 20-year-old was dating a 90-year-old.

It's likely that the 20-year-old would rate that 90-year-old a lot lower than the model would predict here because we built the model from a much narrower set of age gaps. But the model is telling us, Regina, that for most dates involving people within a reasonable age span, the age penalty was actually pretty much negligible. Right.

[Regina] (31:34 - 31:38)
I know. It's interesting, right? It made me feel much better about aging.

[Kristin] (31:39 - 31:45)
So technically, there is a statistically significant effect, but it's so tiny, it's negligible.

[Regina] (31:46 - 32:08)
Right. This is textbook example of statistical significance versus practical significance. And by the way, this is why some statisticians want to get rid of the term statistically significant and replace it with statistically discernible, which I like because sometimes we can discern a difference that is so tiny it's meaningless in real life.

[Kristin] (32:09 - 32:46)
This is such an important point in statistics and something I think a lot of people are not very aware of. We haven't yet done our episode on p-values, Regina, where we'll get into even more depth on this. But here's the key idea.

If you have a huge sample size, that gives you tremendous statistical precision. That means that you can detect even very, very tiny effects, and they may indeed come out to be statistically significant because of the extreme precision, but they can still be so small that they actually don't matter in the real world. And we call this statistical significance versus practical or clinical significance.

[Regina] (32:47 - 32:49)
Yep. Perfect example right here.

[Kristin] (32:49 - 32:55)
I have a statistics article that I wrote that's kind of fun on this topic, and we'll put that in the show notes.

[Regina] (32:56 - 33:13)
Now, the last measure in this analysis was whether the person wanted a second date, yes or no. So it's a binary variable, and I really like this one because it's very concrete. Exactly, and this calls for what is called logistic regression, where the results come out in something known as log odds.

[Kristin] (33:14 - 33:29)
I teach like a whole class on logistic regression, Regina, so I spend a lot of time talking about log odds, and they have beautiful mathematical properties, but they're kind of useless for human interpretation. Even statisticians cannot think in log odds.

[Regina] (33:29 - 33:50)
Exactly. Now, here, Kristin, the researchers did try to translate these log odds into something more interpretable, but their translation wasn't quite right. Think of it like if you're translating something from English to French, but you made some grammar mistakes along the way.

I'll put the details in the show notes about what went wrong here.

[Kristin] (33:51 - 33:54)
Good, and, Regina, did you clean up their translation for them?

[Regina] (33:55 - 34:08)
I did, I did. Since they shared all their data and code, I re-ran the models and converted the results properly into what is known as odds ratios, which are much more familiar. So I'll give one example of the results, and, Kristin, maybe you can walk us through what it means.

[Kristin] (34:09 - 34:31)
When you're looking at the effect of women's age on whether men said they wanted a second date, the odds ratio was 0.99. All right, an odds ratio of 0.99 means that for every one extra year a woman ages, her odds of getting a second date go down by 1%. That's extremely tiny.

[Regina] (34:33 - 34:48)
And just to point out where that 1% comes from, subtract 0.99 from 1, then turn it into a percent. Kristin, you and I have written papers on logistic regression and how to interpret odds ratios in plain English. Let's put a link to those in the show notes, too.

[Kristin] (34:48 - 35:26)
Yeah, let's do that, because odds ratios are tricky. An odds ratio tells you the increase or decrease in the odds of something happening, but people want to interpret them as the increase or decrease in the chance of something happening. But an odds is not the same as a chance.

An odds is the chance of something happening divided by the chance of it not happening. So if you have a 50% chance of getting a second date with your blind date, then your odds are one to one. If you have an 80% chance, then the odds are 80 to 20 or four to one.

So four times out of five, you'll get the second date and one time out of five, you won't.

[Regina] (35:27 - 35:53)
Exactly, so I did convert it to a ratio that lets us talk about the chance of a second date, not odds, and that chance ratio was 0.996, which means that for every extra year a woman ages, her chance of getting a second date drops by about 0.4%. That is what we would call microscopic.

[Kristin] (35:53 - 36:02)
And I just want to remind everybody, we are using a little bit of causal language here, but this is an observational study, so just keep that in the back of your mind. So it's just easier for English.

[Regina] (36:03 - 36:27)
Thank you. And by the way, men and women showed basically the same pattern, no significant difference between the genders. So I like to push the models again.

I calculated how long would it take to drop your chances by 1%, you know, from 99% to 98%. You would need to age two to three years, but to cut them in half, 176 years.

[Kristin] (36:29 - 36:50)
So basically time is not your enemy here, unless you're planning to date for the next two centuries. But again, I've got to remind everybody, we are totally extrapolating from the sample here. I imagine that if you actually paired a 20-year-old man with an 80-year-old woman, that the probability of him asking for a second date is probably quite low and not well-predicted by the model here.

[Regina] (36:51 - 37:13)
Exactly. But Kristin, there were some outliers in these age gaps, and it was kind of fun for me to go through the data and pull out these individual cases and imagine what the dates were like. These people make up stories in my head for what went on.

I pulled out a few to give some anecdotes. Maybe you'd like to hear about those.

[Kristin] (37:13 - 37:14)
Oh, I would love to.

[Regina] (37:15 - 37:33)
Maybe a short break first.

Kristin, we've talked about your Medical Statistics program. It's just a fabulous program available on Stanford Online.

Maybe you can tell listeners a little bit more about it.

[Kristin] (37:34 - 38:11)
It's a three-course sequence. If you really want that deeper dive into statistics, I teach data analysis in R or SAS, probability, and statistical tests, including regression. You can get a Stanford professional certificate as well as CME credit.

You can find a link to this program on our website, normalcurves.com.

Welcome back to Normal Curves. We are talking about age gaps in dating today, and Regina was about to tell us about some fun outliers in the dataset.

[Regina] (38:12 - 38:42)
It was really fun to go in and look at the data and imagine the dates, Kristin. There was this one dude I found who was 54, but he said he would not date a woman over 50. Of course, she's got to be younger, wouldn't even consider someone his age, but then the matchmaker gave him a 61-year-old.

[Kristin]
Ooh, bold.

[Regina]
Very bold. But what happened?

He rated her a four on desire, a five on overall rating, and said he wanted a second date.

[Kristin] (38:43 - 38:46)
Wow, that's impressive. So that matchmaker must be good.

[Regina] (38:46 - 39:15)
Mm-hmm. Hopefully he was surprised and changed his whole idea about aging in women. There was a 28-year-old guy who was paired with a 59-year-old woman.

[Kristin]
Wow. Yikes.

[Regina]
31 years older. More than double his age. His age maximum, he said, was 58. He was okay with 30 years older.

Yeah. No, he opted not to pursue a second date on that one.

[Kristin] (39:15 - 39:17)
Okay, well, that might not have been about age. We don't know.

[Regina] (39:17 - 39:37)
Right, we don't know. There was, however, one 32-year-old guy who got fixed up with two older women, a 56-year-old and a 59-year-old. He wanted a second date with both of them.

[Kristin]
Wow.

[Regina]
And Kristin, I did not know these men existed, he said that his maximum age was 60. He must like older women.

[Kristin] (39:37 - 39:45)
He must like older women, so these men are out there, Regina. I'm telling you, 34 does not seem outside of that gap then.

[Regina] (39:45 - 39:52)
It does not seem that crazy, does it, in comparison? Yeah, compared to that, yeah. So that 31-year-old that I met on the plane?

[Kristin] (39:53 - 39:54)
Oh, yeah, you gotta tell that story, Regina.

[Regina] (39:55 - 40:12)
He was a very lovely guy. We had a lovely conversation about chaos theory and fractals.

[Kristin]
Oh, nerdy.

[Regina]
He asked for my number, suggested we go out for drinks, but Kristin, I don't think he knew how old I was.

[Kristin]
You do not look your age.

[Regina] (40:12 - 40:51)
It was dark. It was dark and the shades were down.

[Kristin]
Maybe he likes older women, just like some of these anecdotes you're telling me about. You should go for it.

I mean, it's 34, 31, all the same, really.

[Regina]
Okay, I'm gonna have him fill out a feedback questionnaire form.

[Kristin]
All right, Regina, you mentioned in these anecdotes that the matchmakers, again, sometimes broke the rules and didn't, impaired them a little bit outside of their comfort zone, and sometimes it actually went fine.

Let's get now to the part where we talk about how did the dates fare when those matchmakers broke the rules and actually went outside of people's preferences?

[Regina] (40:52 - 40:52)
Yes.

[Kristin] (40:52 - 40:53)
How did they analyze those data?

[Regina] (40:53 - 41:23)
Yes, this is great. This is where we're looking at the effect of those age limits. And to test that, Kristin, they used something called discontinuous regression, and they did this for all the outcomes, that romantic attraction, overall rating, chance of a second date, because you can use this both for linear regression and logistic regression.

You can do this discontinuous trick on different types of regression, which is pretty cool. I won't go into the details here. I'll spare you.

[Kristin] (41:23 - 41:28)
We will put some more details in the show notes, though, for our more nerdy listeners.

[Regina] (41:28 - 41:53)
Yes, exactly. But basically, the idea is that you fit one line where the partners were inside the person's age limit and another line for where the partners were outside that age limit, where they broke the rules, and then you check to see if there is a sudden drop or a steeper slope. Are those lines different before and after?

Once you cross that line, do things change?

[Kristin] (41:53 - 42:05)
As in you're 49, great, and then the moment you hit 50, nope, then everything drops down immediately. So what did they find? Were people actually harsher above their stated age limit?

[Regina] (42:05 - 42:31)
They were not, surprisingly. So when they were in their age limit, people rated older dates a little lower, just like we found before, but there was not an extra penalty for your date being over that self-declared age limit. There was no sudden drop, no falling off the cliff, no steeper decline, and no significant difference between men and women.

[Kristin] (42:31 - 42:37)
Oh, that's interesting. So these cutoffs are really not that hard and fast, and the matchmakers realize this, that they're a little flexible.

[Regina] (42:38 - 42:59)
They are, which makes sense, right? Because if your age limit is 50, is a 51-year-old really that much different than a 50-year-old?

[Kristin]
Yeah, probably not.

[Regina]
But keep in mind, of course, these were not random rule-breaking matches. The matchmakers bent the rules only when they thought the person was a particularly good fit.

[Kristin] (43:00 - 43:11)
Right, so these over-the-line dates were probably stronger matches in other ways, which would soften any age penalty. So these findings may not apply to all dates.

[Regina] (43:12 - 43:23)
Exactly, and most of these dates where the rules were broken were not really that far over the line anyway. On average, they were only about a year older than the cutoff.

[Kristin] (43:23 - 43:36)
Okay, so we have to be a little cautious in interpreting these results. We can't really claim big lessons from this. It doesn't apply to all dates.

It's more like a handful of cautious experiments and not a real stress test of those age limits.

[Regina] (43:36 - 43:40)
Right, but it is useful for Tawkify and their matchmakers.

[Kristin] (43:40 - 43:49)
Oh, yes, so they can tell their matchmakers, go ahead and be a little flexible. You do not have to follow these hard and fast rules. That's very useful for the company, actually.

[Regina] (43:49 - 43:56)
It is very useful. Maybe, you know, don't go 30 years over the limit. That might be pushing it.

But a year or two, not too bad.

[Kristin] (43:57 - 44:01)
Yeah, within reason. All right, Regina, so let's sum up their findings here now.

[Regina] (44:01 - 44:18)
Mm-hmm, both men and women were slightly more attracted to younger dates, but the effect was so tiny, it was almost laughable, as we saw. And the real surprise, I think, is that women's preference for younger men was just as strong as men's preference for younger women.

[Kristin] (44:18 - 44:26)
Which is not what we'd expect. And again, really, really tiny, but it wasn't like women had this huge preference for older men.

[Regina] (44:27 - 45:08)
Okay, those, Kristin, were all of the results. Let's talk about what the authors had in their discussion section, because it was actually quite readable. I enjoyed it.

I recommend reading it. They pointed out the limits of the sample, and they were quite transparent about it. They pointed out that maybe the results showed up, the surprising results about men and women and no difference between them showed up because in this sample, people were willing to hand over control to a matchmaker, which is not typical.

It might mean that they were more open-minded and maybe less locked into this typical older man, younger woman stereotype.

[Kristin] (45:09 - 45:10)
That makes sense. Mm-hmm.

[Regina] (45:10 - 45:48)
They also said, okay, let's assume that it's not just this sample, that these findings apply to a bigger population. They said, okay, if that's true, then why do we mostly see couples with older men out in the wild? And I like what they suggested.

They said, first date attraction is not the whole story. And I love how they put it, that somewhere between the blind date and settling down, those older women, younger man pairings that were so preferred may, quote, wither as the liabilities of men's youth come to the fore.

[Kristin] (45:51 - 45:57)
Young men do have some liabilities, yes. Regina, I think they do also come with certain advantages.

[Regina] (45:59 - 46:08)
Paul and Eli, they did not enumerate the advantages, surprisingly. Okay, we're going to keep it PG-13, not even go there, Kristin, but.

[Kristin] (46:09 - 46:20)
Fair enough, but it does echo, Regina, the wishlist dating study that we talked about before from the same authors. What people say they want doesn't always match what they actually respond to in practice.

[Regina] (46:21 - 46:23)
At least at the first date stage.

[Kristin] (46:23 - 46:41)
Yeah, I mean, the first date is not the same as I'm going to marry you, so there is a big difference there, right? If you ask them again later down in the relationship, maybe age matters more then. That's a possibility, yeah.

All right, Regina, I think we're now ready to wrap up and rate the strength of evidence for our claim today. And Regina, can you repeat our claim?

[Regina] (46:41 - 46:53)
That age matters in dating. Of course, with a caveat that we're not talking about Anna Nicole Smith and her billionaire elderly husband. We're talking about it for normal people.

Age in dating matters.

[Kristin] (46:53 - 47:09)
All right, and how do we rate the strength of evidence for this claim with our one to five highly scientific smooch rating scale? One means little to no evidence for the claim and five means very strong evidence. So, Regina, kiss it or diss it.

Are you giving this a second date?

[Regina] (47:11 - 47:31)
I'm going to give it one kiss. One little goodnight kiss. It doesn't sound to me that these effects are very strong.

They might be there, statistically significant, but they're not big. It sounds like the age limits don't matter and it sounds like even this age penalty is not really huge. One kiss.

You?

[Kristin] (47:32 - 48:29)
You're not inviting it up upstairs.

[Regina]
I am not inviting it upstairs.

[Kristin]
All right, Regina, actually, I'm going to go with three smooches on this one, not completely based on the evidence in this paper because I think the evidence is somewhat limited, right?

They looked at people who were dating within a pretty narrow age range. You know, we didn't have a lot of 28-year-olds paired up with 60-year-olds. And these were couples that were more likely to be successful because they had the help of a human matchmaker.

So I'm not sure this generalizes beyond this very narrow case. Maybe important for Tawkify matchmakers to get more bonuses. But if we're talking about just dating out in the real world, I think age does matter some and there's probably other evidence of that.

So I'm going three smooches because I think we can't extrapolate too far from this paper, not to mummy territory.

[Regina]
Okay, I feel like that's fair.

[Kristin]
Regina, how about methodologic morals? What do you have for us today?

[Regina] (48:29 - 48:38)
Do not be swept off your feet by statistical significance. Tiny effects in bed are still tiny.

[Kristin] (48:40 - 48:51)
That is very appropriate for today's episode. Regina?

[Kristin]
And you?

[Regina]
How about this? Fancy units sound smart, but plain English wins hearts.

[Regina] (48:52 - 48:54)
Oh, I love it. Aw, it wins my heart.

[Kristin] (48:54 - 49:24)
Got to convert those log odds in standard deviation units. Yes.

[Regina]
Yep. Oh, very nice. All right.

[Kristin]
I have learned a lot, Regina, from this episode and it's nice to think that the age penalty perhaps isn't that large. I think this opens up everybody's dating worlds and Regina, I think you've got to now go out and collect data, got to date a 34-year-old or perhaps a 31-year-old and report back. And report back.

One date. One date only.

[Regina] (49:25 - 49:26)
I'll submit the questionnaire. Yes.

[Kristin] (49:28 - 49:29)
All right, next time.

[Regina] (49:29 - 49:30)
Thank you, Kristin.

[Kristin] (49:31 - 49:32)
Thanks, Regina. And thanks everyone for listening.