Don't accidentally mislead yourself with graphs

Don't accidentally mislead yourself with graphs

May 10, 2022
–
Ravi Parikh, Founder at Airplane.dev

Recently, I was looking at this graph of the number of weekly active users using Airplane each week over the last few months of 2021:

Line graph of Airplane weekly active users over time
Y-axis omitted intentionally

To me, it looks like boring, linear growth. We're adding roughly the same amount of new active users each week. But this was really surprising to me. I knew for a fact that we had a big inflection point in growth towards the end of September when a blog post of ours did really well on HN, LinkedIn, etc. It led to a one-time big spike in signups and a sustained higher rate of growth afterwards.

I already knew this before I made the graph, which was one reason I was surprised when it looked linear and decided to dig in a bit further.

Look at the graph again with trendlines overlaid:

Airplane weekly active users graph with trendlines overlaid
Y-axis omitted intentionally

We added more than 4x as many new users each week in October and November as we had in August and September. This is not at all obvious when glancing at the graph without trendlines.

Our ability to perceive differences in slope is weak compared to things like differences in length or position (Cleveland and McGill, 1986). If you're trying to tell whether the rate of something is changing over time, plot the rate directly. Create a graph where each bar or dot represents the change in X from the previous time period, rather than the total count of X in each time period. If we do this with our graph above, we get something more like this:

Change in Airplane weekly active users each week
Y-axis omitted intentionally

In this graph, it's clear that in the first half, the change in active users was fluctuating above and below 0 each week. But after late September, the change in active users was consistently positive each week. It's not clear that it's a 4x difference, but it provides a clear signal to dig in and look at the data directly.

If you're trying to compare two quantities, use data visualization techniques where those quantities are represented by easy-to-compare things like bars, rather than hard-to-compare things like slopes of lines.

Share this article:

Subscribe to new blog posts from Airplane:

A downturn is the perfect time to start a startup

A downturn is the perfect time to start a startup

May 23, 2022
–
Ravi Parikh
Tech is going through a steep market correction right now. Companies are doing layoffs. Y Combinator published a letter to their founders telling them how to prepare for tough times ahead [https://twitter.com/refsrc/status/1527238287471292417]. Counterintuitively, this is a great time to start a new company–even better than in 2021 when tech was red-hot. I have friends who were considering starting a company this year, but are now feeling a bit skittish. This is the exact opposite of how you s
How to evaluate startup offers in a downturn

How to evaluate startup offers in a downturn

May 18, 2022
–
Ravi Parikh
It feels like a precarious time to join a tech startup. In the last few months, public market valuations of companies have come way down. The Bessemer cloud index [https://cloudindex.bvp.com/], which tracks average public SaaS company prices, shows that the total market cap of SaaS companies has been cut by more than half in the last few months. This has been going on for a few months, but in the last few weeks, reality seems to have hit startups all at once–the funding market has basically drie
go to home