Eat, Sleep, Poop: The Math Behind Keeping a Human Alive the First 12 MonthsReading time: 4 minutes
My wife and I did it. We successfully kept our firstborn alive for a full year.
I say this partly in jest, but as a new parent your day is full of constant questioning. Did she eat enough? Is she warm enough? How long has it been since she pooped? Pretty often these questions come at 4 AM when you’re sleep-deprived and your Google navigation isn’t as sharp. So, not only do we have a lot of questions, we also get a lot of misinformation.
Did you know, for instance, that whether or not babies need to be burped is up for debate? In fact, a 2015 study found that burped babies actually spit up more than twice as much as non-burped babies. All of this is to say that everyone has an opinion on how to properly care for your baby, and it is a very confusing time. In order to keep our sanity—and as is our nature—we decided to track some data. Because data don’t have opinions.
What Goes In, Must Come Out
It was impossible to remember all of the feedings and diaper changes in the early weeks. They happen at ALL times. So my wife and I downloaded the Baby Tracker app, which helped us keep track of all of the comings and goings of our daughter. Since that first week, we have continued to use the app, which means that we have a full year of data.
I was able to export the data from the app in CSV format. It contains a lot of information. Feedings, diapers, growth, milestones, etc. We didn’t use all aspects of the app, but diapers and feedings were tracked religiously. Once I had downloaded all of the data, naturally I turned to Construct to help me organize and make sense of it.
I pulled the files into Construct and was able to append all of the diaper and feeding data, then quickly added some additional details to the data. I created transformations to the date fields to include an “hour of day” field and calculated things like how many days old and how many months old at any given date.
Feeding the Baby to Feed the Model
One of the main questions we had during the first couple of months was, how much is she going to want to eat? We were using formula, so we had to prepare bottles—and it turns out formula is expensive. We didn’t want to make a six-ounce bottle if she was only going to eat one ounce. So I used Construct to create some measures around recent feedings. This included variables like time since last feeding, ounces in last feeding, feedings in last 12 hours, ounces in last 12 hours, current hour, etc.
I analyzed this prepared data in Predict to create a predictive model that could help us calculate how many ounces to prepare for the next feeding. Predict gave me the ability to use all of this information to predict how many ounces our daughter would eat in the next feeding. Because the number of ounces should be increasing as she got older, I used a “days since birth” calculation to help explain that increase. This was found to be one of the most predictive measures.
While “days since birth” was considered a top predictor, many of the other variables that I created in Construct also turned out to be quite predictive. For instance, the number of feedings in the last 12 hours, the amount of time since the last feeding, and the number of ounces in the last feeding were all significant. Predict allowed me to perform some k-means clustering on these variables, specifically the “time since last feeding” and the “ounces in last feeding” to find a more meaningful relationship.
Like a Sleep Monitor, But for Your Data
The model was fairly accurate, but I needed a way for my wife to be able to access it. For this, I simply applied the model to the data and then outputted this information to Tableau directly from Construct. This allowed me to create a dashboard to keep track of all of our daughter’s feedings and diaper changes as well as the costs, and I included a worksheet that allowed me to enter a couple of parameters to calculate how many ounces to make for the next bottle.
My wife and I both had access to this dashboard, and because I am not sure how psyched my daughter would be in the future that I shared this information publicly, I haven’t made that available here.
Instead, I am sharing just a snippet of some of the data that I collected. In the dashboard below, I attempted to show how my wife and I were really trying to get on a schedule. As I mentioned earlier, everything is up for debate, and whether you should work towards a schedule or just be on-demand is certainly a contested topic.
Regardless, the data shows that in the first two or three months, feedings happen at all hours. It wasn’t until about the fourth month that we started to hit any regular intervals.
Some of you may also hate this chart, because in comparison our daughter has been very easy on us. She started sleeping through the night pretty much after three months. And between three and six months, we were doing a “dream feed”—a feeding when the child is sleeping, in the hopes that she’s eaten enough to last till morning—which is why you see a lot of feedings at 11 PM. We weened off this feeding in the seventh month and have been going strong with our schedule since.
Year one is in the books and, all things considered, it went very smoothly. I have no idea what we have in store for these next couple of years, but isn’t that pretty much what parenting is all about?