Time Series

A “time series” is a data set in which the data points are shown in concurrent order and at regular intervals.

For example, let’s say we want to track how many jellybeans we eat every day for a month. A time series would show jellybean consumption at daily intervals over the course of that month, i.e., ten jellybeans on the 1st, six on the 2nd, 18 on the 3rd, and so on. The point of time series data is to show patterns and trends, so it’s crucial that two things happen: (1) We need to make sure we're regularly collecting data at the intervals specified, i.e. can’t just skip a day of jellybean counting, and (2) we need to make sure we include enough time intervals to make the data set valid. Like...we can’t just count jellybeans for two days and extrapolate that to the rest of the month. We have to give ourselves enough data points to draw valid conclusions.

Time series data has a bajillion uses in the real world, far beyond just monitoring jellybean consumption. It’s used for everything from charting a stock’s performance to documenting population growth to analyzing the impact of laws and regulations. Basically, if something changes over time and we can put numbers to it, then we can put together a time series. And once we’ve started to identify those patterns and trends—maybe we eat twice as many jellybeans on Mondays as we do on Fridays—then we can start forecasting and making predictive models based on what we’ve learned. And if we don’t like the trends we’re seeing, we can start implementing changes (ask the gf to hide the jellybeans on Sunday night, for example) and monitoring their impact.