In the generation that's having kids now, many have a friend whose birthday is 9/11. And they saw that friend suffer through their teens, never feeling right to party on their birthday. It doesn't surprise me they don't want their kid to experience the same.
You aren't. If you drill into the data, the values for each day is AVG(#birth on 1994-MM-dd, 1995-MM-dd, etc., 2014-MM-dd). For 02-29, the cells for, say, 2001-02-29 doesn't exist, so it's merely AVG(1996-02-29, 2000-02-29, 2004-02-29, 2008-02-29, 2012-02-29). (This is conjecture as to how the researcher computed the numbers from the raw data tables, but the wording of the summary suggests this is the algorithm).
Why not do a frequentist approach compared to the population of all the data collected for birthdays? It would be much more accurate, since 02-29 would then appear to be very much an exception.
This is partly in response to the comment mikeash made, but also a question I've been carrying around for a long time. Mikeash wrote
> 9/11 has a small but noticeable dip compared to the surrounding dates. I wonder if that's people avoiding it or if it's just coincidence
I also see something else: there is a light streak going down the 13th of every month, as if people were having fewer babies that day.
If I want to figure out whether this is an actual pattern or just a coincidence, how would I go about that? I mean, as I understand the general process it's a two-parter:
1. Come up with a model of the situation: say, uniform probability of giving birth any given day.
2. Calculate the probability of getting your actual results based on the model. If these are lower than some threshold you pick, say, 5% or 1%, then you discard your model because you have found something that breaks your assumptions.
But there's lots of nuance here. Is does my model even make sense? Isn't it more useful to define the number of child births on any given day to be normally distributed with mu = average over the year? Are the two equivalent because something something central limit theorem?
Is there a simple and convenient way to do the calculations for the most common models? And say I'm interested only in the streak of the 13th... do I need to consider all other days as well before I discard my model? My intuition says that the streak of the 13th is easier to prove/disprove than the dip of 9/11, because it's 12 data points (is there?) compared to just one -- but I also feel like I'm way off base here...
I'm at a bit of a loss here. I haven't studied a lot of statistics because I've been busy with other things but this is the kind of thing I'd love love love to learn.
I do remember a bit of the Think Bayes book, and I could plug each month into a comouterprogram updating the birth probabilities of each day based on that... but how do I know if I have enough data to get a day-by-day resolving power? Maybe I should look at weekly averages instead?
I mean, I understand this is a whole field of research and not something you learn overnight, but is there not some subset of "countertop hypothesis testing" that you can apply casually over the kitchen table to get you at least somewhere?
TL;DR: I think I see patterns in this data and I want to statistically reject my hypotheses. How do I approach that?
Edit: I guess the reason my question differs slightly from high school stats is because I don't get to design an experiment that will give easily analysable data. And I'm afraid of creatively interpreting the existing data to make it easier to deal with, because I think that will lead me to incorrect conclusions.
Sorry I can't answer your question, but I'd bet money it's not a coincidence.. Scheduled births (whether induced or c-section) are prevalent enough that people will just avoid "yucky" dates even if they're not superstitious.
Anecdotally, I went in with my wife for an ultrasound on a Friday the 13th and it was definitely less crowded that day. Rest assured, no Rosemary's Baby detected, nor any heart defects, etc. I just benefited from not having to take a seat in the waiting room :).
I’s love to know that too. I assume some part of the September bump is “festive cheer”, which would still show in Australia. But it’d be fun to know how much was each.
It's been mentioned in other articles, but the ability to induce birth allows for mothers to avoid birthday on holidays. They can plan, and with assistance, schedule a delivery a day earlier for example.
The February 14 spike is particularly interesting as it’s an example of people _choosing_ a ‘positive’ date, vs just avoiding negative dates (holidays, 9/11, 13ths).
February 29 is still pretty close to the average. It also has a smaller dataset than the other days, making the difference less meaningful. It was only two spots down from January 29, and I doubt people were avoiding that date intentionally.
I find it curious that there's a spike on February 14, but not one on November 14. There's somewhat of an increase starting at November 14, but it's not clear if that's a consequence of Valentine's Day. It's as if planning a birth on the holiday motivates reproduction more than the holiday itself.
Pregnancies aren't exactly 9 months. They're closer to 40 weeks which is November 7th which does have a moderate hump compared to the surrounding week.
There are many births that are planned C-sections for when the mother is unable to deliver vaginally safely. These births would not be planned on a holiday because doctors usually only work emergencies on holidays.
It would be interesting to have some data like that per country. For example in Japan, I would suspect that the first places are in April and May. The fiscal year beginning on the 2nd of April, couples prefer to be able to put a kid to the daycare in its first year in the 0-year-old class (0歳児) with a kid the maximum older. So if a kid is born on the 2nd of April, he will be able to go to the daycare of 0-year class when he is 11 months and 30 days, while if he is born in March, the kid could just enter the daycare in the 1-year class (1歳児), which is way harder to join because there are less new places in 1-year classes than 0-year classes. Note that the kid can just go to the daycare if older than 6 weeks (I may not recall details correctly here), so it is possible for kids born in February to go to the daycare of 0-year class, but most of his/her classmates would likely be closer to turn 1 once they all join the daycare, which makes for a huge difference at this age.
I had always thought that April/May was prime birthday season. So many of my family and friends have birthdays in that time of year. I was very surprised to find that it was not a particularly likely time to have a birthday.
I wonder if this error is common in other people. And if so, if there are social-network dynamics around birthdays that cause the bias.
I work a lot with visualisation. It becomes interesting when you release a chart to end users only to get feedback such as lurcio's.
At first I would simply correct and "educate" the end user how to interpret the charts. Recently though I have begun to "register" such types of questions and misunderstandings. I now actively determine WHY the end user made such a seemingly simple error in usage. And more than likely, I've concluded, it's the fault in the presentation of the chart than it is of the user's understanding of it. It is not easy to create data visualisations that do not mislead!
In this case, we as users are introduced to a gradual shading of a pink/maroon palette of calendar dates. The shading goes from dark (almost black) to light pink (almost white). So both white and black squares are tacitly presented as features. The mind extrapolates this. Using white for an invalid date is bound to create confusion to a user not paying attention. And charts are always sold to end users as "you relax, we've done the thinking, just consume this easy picture". As users, our guard is down by invitation!
I'd make all the invalid dates a colour from a completely different palette or strikingly different colour (eg. gray, yellow or green) to stand out clearly, indicating immediately that you are looking at invalid areas of the chart.
In the UK, babies are more likely to be born on a Thursday as hospitals do what they can to prevent babies being born at the weekend when there are fewer doctors around. No actual data, but that’s what the midwife told us.
Any idea what causes the late November dead zone? People seem to be inducing to avoid a Christmas birth. But what happens on the last week of November?
42 comments
[ 5.2 ms ] story [ 109 ms ] thread> 9/11 has a small but noticeable dip compared to the surrounding dates. I wonder if that's people avoiding it or if it's just coincidence
I also see something else: there is a light streak going down the 13th of every month, as if people were having fewer babies that day.
If I want to figure out whether this is an actual pattern or just a coincidence, how would I go about that? I mean, as I understand the general process it's a two-parter:
1. Come up with a model of the situation: say, uniform probability of giving birth any given day.
2. Calculate the probability of getting your actual results based on the model. If these are lower than some threshold you pick, say, 5% or 1%, then you discard your model because you have found something that breaks your assumptions.
But there's lots of nuance here. Is does my model even make sense? Isn't it more useful to define the number of child births on any given day to be normally distributed with mu = average over the year? Are the two equivalent because something something central limit theorem?
Is there a simple and convenient way to do the calculations for the most common models? And say I'm interested only in the streak of the 13th... do I need to consider all other days as well before I discard my model? My intuition says that the streak of the 13th is easier to prove/disprove than the dip of 9/11, because it's 12 data points (is there?) compared to just one -- but I also feel like I'm way off base here...
I'm at a bit of a loss here. I haven't studied a lot of statistics because I've been busy with other things but this is the kind of thing I'd love love love to learn.
I do remember a bit of the Think Bayes book, and I could plug each month into a comouterprogram updating the birth probabilities of each day based on that... but how do I know if I have enough data to get a day-by-day resolving power? Maybe I should look at weekly averages instead?
I mean, I understand this is a whole field of research and not something you learn overnight, but is there not some subset of "countertop hypothesis testing" that you can apply casually over the kitchen table to get you at least somewhere?
TL;DR: I think I see patterns in this data and I want to statistically reject my hypotheses. How do I approach that?
Edit: I guess the reason my question differs slightly from high school stats is because I don't get to design an experiment that will give easily analysable data. And I'm afraid of creatively interpreting the existing data to make it easier to deal with, because I think that will lead me to incorrect conclusions.
> The rate of caesarean section births in the U.S. was 32.7 percent in 2013
With so many c-sections being performed there is more selection of birthdates.
Anecdotally, I went in with my wife for an ultrasound on a Friday the 13th and it was definitely less crowded that day. Rest assured, no Rosemary's Baby detected, nor any heart defects, etc. I just benefited from not having to take a seat in the waiting room :).
Personally, there was a pretty significant blizzard ~40 weeks before I was born. Snowstorm babies unite.
Avoiding holidays was the obvious trend, but there were two others I saw that made me chuckle:
1. People avoiding February 29, so they don’t give their kid that 1-in-4 birthday pain.
2. The spike on February 14! People want their kid born on Valentines (cute, but not really romantic...)
I wonder if this error is common in other people. And if so, if there are social-network dynamics around birthdays that cause the bias.
If you mouse over the individual squares of the image that follows the words:
you will get popups showing more details for each particular date.A sample popup text says:
At first I would simply correct and "educate" the end user how to interpret the charts. Recently though I have begun to "register" such types of questions and misunderstandings. I now actively determine WHY the end user made such a seemingly simple error in usage. And more than likely, I've concluded, it's the fault in the presentation of the chart than it is of the user's understanding of it. It is not easy to create data visualisations that do not mislead!
In this case, we as users are introduced to a gradual shading of a pink/maroon palette of calendar dates. The shading goes from dark (almost black) to light pink (almost white). So both white and black squares are tacitly presented as features. The mind extrapolates this. Using white for an invalid date is bound to create confusion to a user not paying attention. And charts are always sold to end users as "you relax, we've done the thinking, just consume this easy picture". As users, our guard is down by invitation!
I'd make all the invalid dates a colour from a completely different palette or strikingly different colour (eg. gray, yellow or green) to stand out clearly, indicating immediately that you are looking at invalid areas of the chart.
What's confusing the visual are planned birth days.
https://www.bustle.com/articles/93018-can-your-birthday-pred...