Hacker News Reader

Comment by Ntrails | Hacker News Reader

Nobody knows, because no one even agrees how to calculate it.

Divorces/Marriages in a given year is worthless.

Divorces (of people married in 1970) / Marriages (of people married in 1970) gives you a number for that "cohort" but isn't exactly true - there are many years left for them to divorce, and that is generations ago.

Personally I'd say it would be useful to look at divorce rates and marriage rates in terms of overall population and growth and argue whether

    (Divorces / total people married)

was increasing compared to

    (Marriages / total people unmarried)

(I'm sure a statistician can tell me I'm wrong)

View on HN

Except someone can be married and divorced multiple times (even within a single year).

Counting the marriages or the divorces is a mistake. You need to count the people over time, but also not consider people who never get married.

I'd be interested in the "survival rate" of marriages that end in divorce, like they calculate for chronic diseases. What's the 5-year survival rate for a marriage across the population? What's the 10-year rate, or 20-year rate? You could segment such data pretty easily by other factors, like region, ethnicity, age at wedding, etc.

There are a number of studies that gather that kind of data, e.g. http://www.bls.gov/opub/mlr/2013/article/marriage-and-divorc...

> Nobody knows, because no one even agrees how to calculate it.

What's wrong with "percentage of marriages that end in divorce"?

That can be skewed by people who marry and divorce multiple times, making "percentage of marriages that end in divorce" somewhat disconnected from the probability of any one marriage surviving.

But if people do marry and divorce multiple times, wouldn't we want the divorce rate to change with it?

It depends on what one wants the "divorce rate" to mean.

Those are still marriages and divorces. I don't see any skewing.

Whether you were married once and divorced once or married 6 times and divorced 6 times, all the times should be counted.

The important factor is marriages/divorces not people/divorces.

That depends on why you are interested in measuring divorce rates.

Is it though? I think the idea is to see how stable marriages are.

And if some people have some "social dicease" that makes them marry and divorce more than others, I'd expect them to be outliers and statistical noise, compared to people that legitimate try to make it work and fail.

Since nobody can have a negative number of divorces, a small percentage of data points with many divorces can "skew" the statistics.

My point is that "a small percentage of data points with many divorces" will not skew the statistics much, precisely because it's small.

So we don't need some with 'negative number of divorces' to balance it, we just have some error bars that don't take away from the average numbers.

That's it if we did want to exclude those people in the first place. But why? A marriage is a marriage. If their 3rd marriage cannot hold, it's not that different than someone marrying once and divorcing.

But you don't know how small the effect is unless you measure!

If 90% of the population marry once and never divorce but the remaining 10% marry and then divorce 5 times this means that the divorce rate is 36%.

The article mentions that divorce rate is usually quoted to support some political platform rather than for it's own sake. For example a politician talking about the breakdown of the nuclear family. If you want to measure how divorce affects nuclear families then you should include only marriages that produced offspring. You should probably also exclude divorces that happened after the children left home. You might get a very different stat.

>But you don't know how small the effect is unless you measure!

The very final actual number no.

But as with most things in life we can have a sense (based on what you know and see around us in society) of it's scale.

In this case, I don't think "percentage of people with multiple marriages" is such a difficult and exotic subject to consider that our guesses will be that far off them mark.

>If 90% of the population marry once and never divorce but the remaining 10% marry and then divorce 5 times this means that the divorce rate is 36%.

That's only if the calculation is based on naive average. You can get percentiles, medians, etc.

"90% of the population marry once and never divorce" is already a valid metric.

I think you just explained why the simple ratio: (# of divorces / # of marriages) is prone to difficulty. The ratio is artificially skewed by people who marry/divorce frequently. So the simple ratio may not actually tell you much about how stable marriages are.

Give the two numbers (#divorce/#marriages) and (#divorced/married_individuals). Some stats may not be summarized in one number only.

Actually, it is very much connected to the probability of any one randomly selected marriage failing. Any specific marriage, of course, you'll want to look at the rates of similarly-situated marriages (the same sources that give numbers in the neighborhood fo 50% for the overall divorce rate often report different rates for, e.g., first vs. subsequent marriages, and those that I've seen have consistently had lower divorce rates for first marriages and increasing divorce rates for each subsequent marriage.)

It doesn't make sense to exclude some marriages from something and calling it the overall divorce rate, even if there are cases where the most interesting divorce rate isn't the overall rate.

You don't know if people will divorce or not until 1) they get a divorce or 2) one of them dies instead. As a result, we can only reliably calculate this percentage for the cohort of couples who married 50+ years ago. We have many reasons to assume that couples marrying now, or who married ten years ago, will not divorce at the same rate as those who married 50 years ago. As another poster says, we can estimate based on historical comparisons and other data. The NY Times article referenced in the link shows some of that: http://www.nytimes.com/2014/12/04/upshot/how-we-know-the-div.... But we can't measure it directly. And we may get surprised - the number of people currently divorcing in their 50s and older is at a historic high, which wouldn't have been predicted by comparisons to previous cohorts.

Ha, this is exactly the issue we ran into at my start-up when trying to determine KPI's.

My start-up is a marketplace that connects coaches and athletes, so we wanted KPI's like "What % of clients rebook their coach?".

But the issue there is the same as marriage stats: You can't just do a simple (# of rebookings / # of bookings), because that will get skewed if, say, are marketing push generates a ton of new bookings.

So then you think, "okay, the denominator should be first-time bookings and the numerator should be how many of those have a later re-booking?" But then you need to give that a chance to happen, because what if someone comes back a year or two later to re-book? Conversely, recent bookings won't have had a chance for the rebooking to happen, so again a wave of brand new bookings will skew the number unfairly.

Ultimately, I think we settled on the "right" decision: cap the time frame, and frame it not as a "rebooking rate" but a "1 month rebooking rate": i.e., "what percentage of new bookings have a re-booking within 1 month?" That one will always generate a correct number, but the disadvantage is your cohort is at least one month old.

So, the same issue applies here with divorce. The only accurate number you can calculate is something like "What is the 20-year divorce rate?" But then you're looking at marriages that occurred a generation ago, so it's not very predictive of marriages that just happened today.

I think the main meaningful number is the chance of divorce for marriages starting today. There are probably many ways to estimate it indirectly.

marriage ends in either death or divorce, surely they can combine the stats!

Reply on news.ycombinator.com