I'd be curious to see what it looks like to plot the ratio of the two lines. That'd capture the relationship trend in a single line. Of the three you provided, I definitely like the last most. The trick, as you mention, is making sure casual readers understand it's a counterfactual. Also making sure they understand the assumption being made is critical.
Oh yes, I was only focused on the period where data for both existed. My assumption was the prior data for Renminbi was missing/unknown, not zero so ignoring it may make sense in this situation. It's definitely a major limitation.
Regardless of whether this actually resolved the issue, you get points for looking up the RGB/Hex color codes for the FT chart and matching them exactly in Stata. Job well done!
Another good point about going until the marginal cost exceeds the marginal benefit: I didn't match the text color. At that point I figured it looked good enough.
I've always struggled with dual axis charts because of what you're showing here. I like your final product, and wonder if a stacked area chart would do the trick or even a proportional stacked one with the 3rd category being "all other currencies"
Stacked area charts are tough. I think they look really cool, but they're basically time series pie charts, which means they carry the flaws of pie charts over multiple years. I think for it to work you have to have each subgroup following a distinct trend over the entire sample period.
That's a great description to their failings... "basically time series pie charts"... I'm going to steal that phrase. I have stakeholders asking for stacked charts all the time when multiple lines conveys a much clearer story. They only get it when I show them both versions.
Yeah, I had a project where I was deriving some conclusions from a stacked area chart, but when I broke it into multiple lines it was easy to see that I was wrong. I've seen cases where they work well, but you have to have a specific type of compositional trend.
Great post Craig. I see this a lot (but missed the one in X which I've now abandoned). Analytical policing is underrated.
I'd be curious to see what it looks like to plot the ratio of the two lines. That'd capture the relationship trend in a single line. Of the three you provided, I definitely like the last most. The trick, as you mention, is making sure casual readers understand it's a counterfactual. Also making sure they understand the assumption being made is critical.
The ratio would make sense, except the Renminbi value is 0 before 2016. So that section of the line is undefined.
Isn't a ratio more volaile if the numerator and denominator are negatively correlated?
Oh yes, I was only focused on the period where data for both existed. My assumption was the prior data for Renminbi was missing/unknown, not zero so ignoring it may make sense in this situation. It's definitely a major limitation.
Regardless of whether this actually resolved the issue, you get points for looking up the RGB/Hex color codes for the FT chart and matching them exactly in Stata. Job well done!
Another good point about going until the marginal cost exceeds the marginal benefit: I didn't match the text color. At that point I figured it looked good enough.
I've always struggled with dual axis charts because of what you're showing here. I like your final product, and wonder if a stacked area chart would do the trick or even a proportional stacked one with the 3rd category being "all other currencies"
Stacked area charts are tough. I think they look really cool, but they're basically time series pie charts, which means they carry the flaws of pie charts over multiple years. I think for it to work you have to have each subgroup following a distinct trend over the entire sample period.
That's a great description to their failings... "basically time series pie charts"... I'm going to steal that phrase. I have stakeholders asking for stacked charts all the time when multiple lines conveys a much clearer story. They only get it when I show them both versions.
Yeah, I had a project where I was deriving some conclusions from a stacked area chart, but when I broke it into multiple lines it was easy to see that I was wrong. I've seen cases where they work well, but you have to have a specific type of compositional trend.