Hacker News Reader

Comment by Cshelton | Hacker News Reader

Ah! Double Y axis charts!

I have been arguing with another founder about the double-y and how we should not use it.

What are peoples' response to the charts in this post? Do you find them easy to read/helpful/giving good insight into the data?

View on HN

I think it depends very much on what you're trying to show. In the article the point he is making is that volume and average price have a roughly inverse relationship.

As with almost all data vis it can be misread but for showing relationships between metrics over time it's the best chart.

I'm not sure how much you should count my input, but this was my first time to have seen a double-y axis chart and I'm still not sure if I was reading it correctly.

Don't worry, that's generally how most people feel, even if you've seen them many times. You have to pause and say, "...wait". I'd argue that the trends you actually see in almost all of them are misguided.

I'd say the bigger concern is spurious correlation. The two factors may very well be unrelated, or share a common cause (like drowning and ice cream prices -- they're both driven primarily by heat).

However, when talking about price and revenue it is legitimate to render them on the same chart, since revenue is a function of price. Steam in particular has proven to be a wonderful platform to demonstrate this phenomenon, as the elasticity of demand for games software is very high, and discounting even modestly generates far more revenue than you lose through the discount. Steam and the publishers benefit even more due to the low marginal cost of delivery, largely accountable to transmission costs and the marginal wear-and-tear on data servers.

I've always liked double y-axis charts. I do wish they'd colour-code the axis labels to match their associated lines.

I think the problem is that there are relationships that are very difficult to show cleanly using other chart types.

Relationships between quantities with different units are hard to show otherwise, without significant cognitive overhead (a new unit ), or two charts which causes inevitable back and forth examination.

You mention elsewhere that they are commonly misinterpreted or misleading. Care to show some examples? I can't think of any off the top of my head.

These particular charts are poorly done because their isn't enough information about which axis refers to which plot line, but are otherwise clear in intent.

I find I have to think about them too long, and I think the rendering details contribute to that. There is nothing wrong with double y-axis graphs if you don't just slap them together because the x-axis has the same range. I think the lack of concern for the rendering details make these harder to immediately read than they should be.

Why are the axis labels not colored the same way the lines are, or the entries in the key not positionally associated with their axis (rather than being in the center). The triangles and circles on the lines on "Steam Sales" offer no utility (except, perhaps, for color-blindness or if it was printed in black-and-white, but then dotted-vs-solid would be a better choice). There are horizontal tic marks, but they are separating the dates, not centered on the datapoint, meaning the data is being presented as a column, which a line graph is not best for (that is, the choice of graph given this data may not be the best one).

And that's just the problem with one graph. There are two graphs on this page, and both have different rendering details. Consistency is important because it means understanding the rendering of one graph translates to the understanding of the other. Why doesn't the second graph have shapes over the data points? Why doesn't the second graph have tic marks along the x-axis? Color is used inconsistently: the unit for the blue line is units sold in the first one and dollars in the second one, and the side the graph the blue line is associated with switches. Despite that the axes that are denominated in dollars are showing different things (totals vs discrete values), there is value in having them both on the same side of the graph (and in the same color)—that the blue one switches sides means the viewer needs to concentrate harder on the key and layout leaving less time for interpreting the data. It would actually be better if four colors between the two graphs were used since there are four different domains of data being shown. They both show the same time range, but the dates are not formatted similarly along the x-axis.

In both cases, the units on the labels are too exact. "60.00%": why is this 4 significant digits, a whole number percentage would do. At this size, the decimal point and percent sign almost get lost with the unneeded zeros. "Steam Sales" has units that are so big with so many zeros you have to think about the order of magnitude; these should be truncated to "$300M" and "65M".

Yes. But I think the problem is they can give misleading insights if you're not careful.

Exactly my argument. I would rather not give the user the option of creating that chart in the first place because 95% of them are misused and give misguiding trends, etc.

If I was creating charts I would want the freedom to do this. Perhaps display a warning if both Y-axes are the same units (e.g. both dollars) because there is little reason to do that - either the values are related and the graph will be hard to parse/misleading, or the values are not related and they probably shouldn't be on the same chart.

But I can certainly see a use case for $ v. %, etc.

> display a warning if both Y-axes are the same units because there is little reason to do so

OK - let's see a line chart showing the correlation between average daily lunch cost and annual salary :)

A perfect example of "I know what I'm doing, let me do something that is usually disingenuous." :)

Reply on news.ycombinator.com