1
$\begingroup$

I have 4 series/variables over time. They are in totally different units (monetary, quantities, elasticities, and percentages). I want to make a line plot for each of them. I know it makes no sense to put in the same graph, not even when using different axes, as they are messy.

What troubles me the most is the selection of the scale of the y-axis (x-axis is time). You can zoom in or out and get visually different degrees of apparent variability of the series (and careless readers might be fooled by this).

Of course, that's why we care about the standard deviation. But still, I want to make a graph of the series alone that are somewhat informative.

My question is: is there some logical scale at which the y-axis is to be presented? For example, do I select the mean and then select the y-axis to cover 2 standard deviations up and down from that mean? Is there some standard here?

PS: I've seen this, this, and this question, but I find no answer to mine there. For instance, the last one, suggested by Nick Cox, has all variables in exactly the same units. In my case, they have for different units. The suggestion of percentage change is not helpful either, because some variables are known to vary much more than others. Ideally, I want to plot the series themselves, and not the changes.

$\endgroup$
4
  • $\begingroup$ Your second and third links are to the same thread. I disagree: the essence of any good answer is contained in those threads. Either plot separately using different scales appropriate to each range, or plot together on some scaled or standardized scale. There can't be a one-answer-fits-all reply, as scaling using min and max, by mean and SD, by median and IQR, etc. might all be sensible. Yet further, as touched upon in the cited threads, transformation might be a good idea too. $\endgroup$
    – Nick Cox
    Commented Mar 11, 2019 at 18:40
  • $\begingroup$ @NickCox That same question was my third link! I erred in the copy/pasting. Added some comment. $\endgroup$
    – laetizta
    Commented Mar 12, 2019 at 8:03
  • $\begingroup$ Sorry, but same reaction from me to your edit. The threads cited do discuss variables on different scales. What is precisely appropriate to your particular data? We can't comment, as you don't show them. What is generally appropriate? Practices vary between using some kind of scaling and just showing variables separately on their original scales, and for good reason, as there are advantages and disadvantages to both. $\endgroup$
    – Nick Cox
    Commented Mar 12, 2019 at 8:13
  • $\begingroup$ Procedure here only allows specification of one duplicate in a vote to close, but your citations imply that the issue is well discussed in several threads, and I can't see that you have a new question. Voting to close as unclear what you're asking was also on my mind. That will all sound negative: FWIW, I am intensely interested in visualization or graphics, and I would post an answer if I had something to add to existing discussions. $\endgroup$
    – Nick Cox
    Commented Mar 12, 2019 at 8:16

0

Browse other questions tagged or ask your own question.