Open Vis Conf 2016: Everything is Seasonal
In April 2016, I spoke to an audience of 500+ people at OpenVisConf on the topic "Everything is Seasonal."
Abstract: People, and our data, are heavily influenced by our regular hourly, daily, weekly, seasonal and annual patterns, as well as by typical (holidays, weather variation) and one-off (natural disasters, electrical outages, war...) aberrations to these patterns.
Time series analysis must take seasonality and seasonal variation into consideration, but so should any analysis comparing data at two different points in time. For example, frustrated with how much worse the traffic has gotten from July to November? Worried it's just going to keep getting worse? Consider that most commuters take a week's vacation during the 8 weeks of summer, but almost nobody takes vacation in early November. Based on that alone, you'd expect there to be around 14% more cars on the road in early November than in July.
This talk uses examples to illustrate 6 key recommendations for thinking about the role of seasonality in your data, both to avoid (sometimes surprising) pitfalls and to reveal insights that are too often aggregated away.