4 ms·
On the dog weight and trade deficit charts, I'd argue there's merit to the originals. The essence of a linear relationship between two data series X and Y is th
by nerdponx 8y ago
On the dog weight and trade deficit charts, I'd argue there's merit to the originals. The essence of a linear relationship between two data series X and Y is that X and Y are identical up to position and scale. The "better" charts represent the data more accurately, but the original charts make it easier to see the relationship between the two series.
Also the trade deficit chart is confusing in part because the negative sign isn't obviously a negative sign.
Edit: the last chart is fine. It's maybe not very sexy, and it takes a moment to absorb, but the data is clearly displayed.
- kbenson 8y agoThe dog weight one is a good case of how you can futz with numbers to imply a greater relationship than is present. I think the second chart is much better in that respect. If you're doubling the relative range of one of a double scale to make stuff look like it fits, that's not exactly being forthright. > but the original charts make it easier to see the relationship between the two series. Well, as the article says, if it looks like they're a perfect fit when they definitely aren't, that's probably beyond making something clearer and going into convincing people of something that the chart doesn't really imply. > the last chart is fine. It's maybe not very sexy, and it takes a moment to absorb, but the data is clearly displayed. Totally agree on this. It takes a bit to digest, but that's because there's a lot of information imparted.
- function_seven 8y agoThere's no reason the scale needs to be the same when comparing measurements in totally different dimensions. The weight is roughly a volume measurement while the neck size is a linear one. Also, if the necks were measured in inches, you'd end up at roughly the original chart, even being mindful of the scaling on each axis. I think the left chart is actually the better one. It does a good job showing the correlation between neck size and weight.
- Quenty 8y agoIt’s a matter of percent change. It shows the same relative percent change on the scale, which feels more accurate. I prefer the revised graph.
- gowld 8y ago> > The weight is roughly a volume measurement while the neck size is a linear one. so "percent change" is a different dimension in the two data sets, respectively.
- ovi256 8y ago>The dog weight one is a good case of how you can futz with numbers to imply a greater relationship What they did was very similar to standardizing data series (normalizing to mean=0 and standard deviation=1). This is a standard tool in scientific data analysis (most published papers use it) and it's taught in statistic and data science classes. It's also a preliminary step in computing most correlation measures. It does not exagerate relationships, it helps visualizing them. The second chart removed the standardization and obfuscated the relationship.
- SiempreViernes 8y agoAs you can clearly see from the axis labels, they are not plotting standardised or indexed data, what they did was hand picked a rescaling that makes the plots line up very nicely. Further, it is simply not true that weight is a perfect function of neck size, so you shouldn't make a plot that implies that.
- gowld 8y agoThe reason it's hard to pick which chart is better is that both charts are far too detailed for the tiny amount of information they convey, which is "both values shrank over time"
- SiempreViernes 8y agoCan you further explain the frankly astonishing statement that you find it hard to pick between a plot showing a false equivalence and one that doesn't?
- ovi256 8y ago>what they did was hand picked a rescaling that makes the plots line up very nicely I bet dollars to donuts that's what a stardization would show too, unless I'm missing something. It would show an even more perfect lining up because it would line up the line ends. Wish I could test it. Wish pandas had a read_image function to get data from line plots like it has read_html.
- soVeryTired 8y agoAs you suggest, it's not always incorrect to shift and rescale data for visualisation. In fact, that is exactly what a linear correlation calculation does: the data is shifted to have mean zero and rescaled to have variance one, then the covariance is computed. The real problem is that often, the two timeseries being compared are too simple to demonstrate whether there is a relationship present. If they're both essentially straight lines as in the Butcher's Dog graphs, you may see a spurious relationship. What you need -- in unscientific terms -- is to see one timeseries do something weird, and the other timeseries reacting to the weird thing in a manner that fits the narrative. Straight lines don't really do that, but often the first difference of two timeseries that look like a straight line will be good enough.
- ukoki 8y agoI don't get the dog one at all. The article is about dogs getting smaller. Since neck size and weight are very correlated why bother to show both? I would just have a single plot showing the decline in dog weight over time. The fact that smaller dogs also have smaller necks is neither surprising nor useful.