1
$\begingroup$

I'm very new to Data Viz and R. I'm trying to understand how graphs are interpreted.

Does the below graph look nice? What would you interpret looking at it? Would you consider there is a slight increase in the Spend?

Data Description: HSHD_COMPOSITION - household composition, describes how many adults, children present in the house, SPEND - How much a particular kind of household spends at a super market WEEK_NUM - week number

I'm trying to understand how each household composition type spends over a span of an year

enter image description here

$\endgroup$
  • 7
    $\begingroup$ It's hard to interpret a plot without context. What is money being spent on? What is the unit of SPEND? What does HSHD stand for? What question are you trying to answer in the first place? Are people multiple-counted if they appear in more than one week? Etc. (I presume of course that by "look nice" you mean how intelligible it is as opposed to how aesthetically pleasing it is, since this site is about data analysis, not graphic design. You won't find many art majors here.) $\endgroup$ – Kodiologist Dec 9 '18 at 15:32
5
$\begingroup$

It's not a bad plot. One can readily see that spending varies quite a bit by household composition, and also varies over time, but the spending of the compositions relative to each other is similar throughout the year.

Here are some suggestions for improvements:

  • Label the x-axis with real dates. Then you can see how peaks and valleys are related to the calendar. Perhaps the spike on the far right is from Christmas shopping.
  • Instead of stacking the household compositions, plot them separately, either as six lines on one plot or as six plots. This will make it easier to see e.g. how much single females spend in total, or how single female spending relates to single male spending, and it will avoid letting the shape for one composition be affected by the shape of compositions stacked below it.
$\endgroup$
  • 2
    $\begingroup$ One could perhaps add that using colour makes plots hard for people with colour vision anomalies. $\endgroup$ – mdewey Dec 9 '18 at 16:43
  • $\begingroup$ I'd agree in that stacking is the main problem with the graph. $\endgroup$ – E. Sommer Dec 11 '18 at 11:48
4
$\begingroup$

I don't think it's a good plot. Why not?

First, the categories of household composition are not clear. In particular, how is "one adult" different from "single female" or "single male"?

Second, it's pretty clear that this is two years. But the years should be stacked. Surely what is spent on a particular day in one year is similar to the same day in another year? If you stacked it, then the apparent peak just before Christmas and sharp drop after Christmas would be clearer.

Third, as @Kodiologist noted, the labels for weeks aren't great. People don't think of "week 22" they think of "June". It should be labeled in months, with labels for, I think, the first of every other month.

Fourth, I agree with @Kodiologist that the stacked nature of the plot isn't great. What if, one week, "single female" drops below "single male"? Also, as @mdewey pointed out, the only cue to category is color, which can be hard for some people. I would use lines (with white space in between) and vary both line color and line type (dash, dot, dot-dash, etc).

Fifth, what is the unit of "spend"? I can't think of one that makes sense here. It should be dollars (or other currency) per week. But right now, I don't know what it is. For money variables, I also like to use a log scale.

Finally, is the data correct? How can "one adult" spend more than "two adults with kids" and also more than "single male" or "single female"?

$\endgroup$
  • $\begingroup$ In the same spirit as this careful discussion: Even tiny details can mar a plot. Here, and everywhere, the variable or column names you use inside a program such as SPEND and HSLD_COMPOSITION should always be replaced by ordinary language for the benefit of readers. Never use all capitals in explanation, so Household composition is better. Note how one experienced analyst couldn't make sense of HSLD. Graphs should not be guessing games. $\endgroup$ – Nick Cox Dec 11 '18 at 11:52
  • $\begingroup$ Should be HSHD not HSLD. $\endgroup$ – Nick Cox Dec 11 '18 at 12:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.