XWe have detected your location as outside the U.S/Canada, if you think this is wrong, you can choose your location.

Macmillan Higher Education Celebrating 20 years of Macmillan Study Skills

Cart

Continue Shopping
All prices are shown excluding Tax
The submitted promocode is invalid
Discount code already used. It can only be used once.
* Applied promocode: ×

Important information on your ebook order

MIHE Blog News, views and insights from Macmillan International Higher Education

wedding-photo

The Probability Of Being The Youngest Person At A Wedding

by Steven Skiena 30th April 2019

Data science is about rooting out the genuine phenomenon underlying extraordinary events, writes Steven Skiena.

It had been a beautiful wedding. We were very happy for Rachel and David, the bride and groom. I had eaten like a king, danced with my lovely wife, and was enjoying a warm post-prime rib glow when it hit me that something was off. I looked around the room and did a double-take. Somehow, for the first time in many years, I had become younger than most of the people in the crowd.

This may not seem like a big deal to you, but that is because you the reader probably are younger than most people in many settings. But trust me, there will come a time when you notice such things. I remember when I first realized that I had been attending college at the time when most of my students were being born. Then they started being born when I was in graduate school. Today's college students were not only born after I became a professor, but after I got tenure here. So how could I be younger than most of the people at this wedding?

There were two possibilities. Either it was by chance that so many older people entered the room, or there was a reason explaining this phenomenon, This is why statistical significance tests and p-values were invented, to aid in distinguishing something from nothing.

So what was the probability that I, then at age 54, would have been younger than most of the 251 people at Rachel's wedding? According to Wolfram Alpha (more precisely, the 2008--2012 American Community Survey five-year estimates), there were 309.1 million people in the United States, of whom 77.1 million were age 55 or older. Almost exactly 25% of the population is older than I am as I write these words. The probability that the majority of 251 randomly selected Americans would be older than 55 years is thus given by:
skiena-sum1
This probability is impossibly small, comparable to pulling a fair coin out of your pocket and having it come up heads 56 times in a row. This could not have been the result of a chance event. There had to be a reason why I was junior to most of this crowd, and the answer wasn't that I was getting any younger.

When I asked Rachel about it, she mentioned that, for budgetary reasons, they decided against inviting children to the wedding. This seemed like it could be a reasonable explanation. After all, this rule excluded 73.9 million people under the age of eighteen from attending the wedding, thus saving billions of dollars over what it would have cost to invite them all. The fraction f of people younger than me who are not children works out to f = 1 - (77.1/(309.1-73.9)) = 0.672. This is substantially larger than 0.5, however. The probability of my being younger than the median in a random sample drawn from this cohort is:
skiena-sum2
Although this is much larger than the previous p-value, it is still impossibly small: akin to tossing off 27 straight heads on your fair coin. Just forbidding children was not nearly powerful enough to make me young again.

I went back to Rachel and made her fess up. It turns out her mother had an unusually large number of cousins growing up, and she was exceptionally good at keeping in touch with all of them. Recall Einstein's Theory of Relativity, where E=mc^2 denotes that everyone is my mother's cousin, twice removed. All of these cousins were invited to the wedding. With Rachel's family outpopulating the groom's unusually tiny clan, this cohort of senior cousins came to dominate the dance floor.

Indeed, we can compute the number of the older-cousins (c) that must be invited to yield a 50/50 chance that I would be younger than the median guest, assuming the rest of the 251 guests were selected at random. It turns out that c=65 single cousins (or 32.5 married pairs) suffice, once the children have been excluded (f=0.672).

The moral here is that it is important to compute the probability of any interesting observation before declaring it to be a miracle. Never stop with a partial explanation, if it does not reduce the surprise to plausible levels. There likely is a genuine phenomenon underlying any sufficiently rare event, and ferreting out what it is makes data science exciting.
There likely is a genuine phenomenon underlying any sufficiently rare event, and ferreting out what it is makes data science exciting.
This is an example of a Data Science war story from my new book The Data Science Design Manual. It is designed to illustrate the power of statistical significance testing to help understand whether an observed phenomenon is a real thing, or just a coincidence. A critical thing in data analysis is determining whether an observation in a given data set or circumstance reflects signal or noise. What do you believe and why? Data Science is much more than programing. Statistical significance testing is the way to demonstrate whether something you see is real, both to yourself and to others.
Featured image credit: Photo by Sweet Ice Cream Photography. Available on Unsplash via the Unsplash License.