Hacker News Reader

Comment by birken | Hacker News Reader

A good data analysis lesson here: Make sure to use multiple Y axes if you are compared line graphs like this.

It is much easier to see the differences at the top of the scale than at the bottom of the scale, but that doesn't mean the shapes of the lines aren't similar if scaled properly. In fact somebody in the comments has made a version with the Y axes normalized, and the shape of the graphs are all pretty much the same, completely contradicting the conclusion the author makes (or if not, the author didn't provide enough clarity on the point he was trying to make).

Classic example of "not challenging your own data" [1]. A tough skill to learn when doing data analysis, because it isn't fun spending a lot of time and effort to try to prove yourself wrong.

1: http://danbirken.com/statistics/2013/11/19/ways-to-make-fake...

View on HN

The comment is here if you're curious about this (as I was):

http://avc.com/2015/06/two-charts/#comment-2108986922

Direct link to the comparison chart:

http://a.disquscdn.com/uploads/mediaembed/images/2217/5785/o...

Great analysis. A post I wrote recently[1], called "Use percentages" reinforces exactly why your post and Tim's comment are important. Absolute values, especially where some are growing more exponentially than others, will always mislead the human eye.

[1] http://veekaybee.github.io/use-percentages/

Another flaw: The Y-axis should be log scale.

In any stock market chart for 5+ years, the Y axis should be log-scale. The stock market tends to go up exponentially over time. If you don't use a log scale, then old prices look flat while recent changes are distorted.

(^DJI interactive chart max time interval)

http://finance.yahoo.com/echarts?s=^DJI+Interactive#{%22rang...

Notice how, on the linear scale, the 1987 crash looks like nothing, but on the log scale, it's obviously one of the biggest dips.

The OP makes the same error. Private equity valuations go up exponentially over time for the same reason that the stock market goes up exponentially over time. Since his time horizon is more than 5 years, the Y-axis should be log scale. He's artificially exaggerating more recent gains by using a linear scale.

Also, with a log scale, he wouldn't need to use a different chart for seed/A/B/C/D.

Now I wonder how sensitive that plot is to the choice of the starting year, since everything is measured relative to itself at 2006.

Can you apply this sort of thinking to taxi rates in the New Orleans metropolitan area?

yeah, without percentages it all looks the same . Series A could be the same percentage as B

Reply on news.ycombinator.com