jasp-stats
[Bug]: Unwanted scientific notation in p values despite preference setting
JASP Version
0.19.3
Commit ID
No response
JASP Module
T-Tests
What analysis are you seeing the problem on?
No response
What OS are you seeing the problem on?
macOS Silicon
Bug Description
Getting unwanted scientific notation for p despite preferences setting values to two digits in preferences
Expected Behaviour
Preferences should be honored ;) or at least use the <.001 convention for very small values of p more consistently. I know scientific notation but any opportunity for error is undesirable!
Steps to Reproduce
Run t-tests after setting preferences?
Log (if any)
No response
More Debug Information
No response
Final Checklist
- [x] I have included a screenshot showcasing the issue, if possible.
- [ ] I have included a JASP file (zipped) or data file that causes the crash/bug, if applicable.
- [x] I have accurately described the bug, and steps to reproduce it.
@boutinb I think I have seen that too. p-values are somehow treated special / separately. But the options do not make that clear enough nor does the help file. And it is also not obvious to the user why they are treated special in the first place. I agree, that often it is good to see a higher precision p-value, while other values have a smaller amount of decimals, but at least the scientific notation should not appear in this case - should it?
I think it might be a bug, since I did notice that later, on a different machine (still MacOS M-series), same type of analysis, but it showed p properly and rounded everything else. (Crosstabs, I mean contingency tables, can be pain with two decimal places on the percentages.)
I’ve never seen scientific notation on p before.
Thanks for taking up this report so quickly!!
I know that there was a lot of discussion about it. @vandenman @EJWagenmakers ?
@boutinb @JorisGoosen @EJWagenmakers @vandenman I think this issue arises when a p-value is greater than 0.001 (so that way of displaying cannot be used), and then the number decimals gets limited. Then, JASP will ignore the option for exact p-values. I think the option is not necessarily about using the scientific notation or not, but about how p-values below 0.001 ought to be displayed. Still, the behavior seems a little weird and I'm not sure it's intentional. Do you think the p-values in the screenshot below should be displayed as 0.1, 0, and 0.1?
I actually found this issue because I received a question during yesterday's JASP workshop, about whether the number of decimals that p-values get displayed with, should be affected by the fix decimals option. For instance, when set to 2, p-values could be listed as <0.01, and <0.1 for 1 decimal, etc. Is this an easy to implement feature request? And I think that also solves the problem described here right?
Steps to reproduce:
- use debug data, go to one-sample t-test
- specify contNormal
- set number of decimals to 1 or 2
1 decimal
2 decimal
That's how I'd expect it to show up with fixed decimal ranges.
I'm guessing this doesn't happen if you select "display exact p-values"?
I think SPSS does it as <.00 which is obviously wrong on the face of it... one way around that would be going as far through 0s as needed to get to a non-zero number, but it would be awkward to see it as .0003 in a column of .08 and such! On the other hand that would be hard to misinterpret.
Seeing the table options again I'm thinking the issue is user error in not displaying exact p-values. This would explain why I saw different things on different computers.
Haha yes we can definitely improve on how SPSS displays those.. so perhaps we can turn this bug report into a feature request where the decimals of the p-value display are affected by the setting too?
Feature request makes sense.
I thought what you describe was what was happening. I think it makes sense to set p separately from other numbers. I know it's a mess to do it that way. Thing is, most of the time if I'm comparing averages, I don't even want decimals, but for p, I generally want two places. I mean, for me, the question is nearly always “below .01,” or “below .05.” (Those who use Bonferroni method and such need more precision.)
PS> Thank you for helping to make JASP largely superior to SPSS wherever I've looked at both.
I'm looking forward someday to automatic Welch's t-tests when the group variances are different! (SPSS' output used to confuse the !@#&*^! out of my students.)
Alright, so currently I get this for pvalues:
I guess having 0 decimals doesnt make the most sense but if people dont want that they should probably not set that :p
Edit: if locale is set to usa english it would have a . as a decimal separator of course.
@JorisGoosen Very nice that this gets improved :)
Also note the discussion on this topic of this duplicate I just closed here: https://github.com/jasp-stats/jasp-issues/issues/3474 Maybe there are additional infos to consider in there.
That looks great already @JorisGoosen!
One tweak that I would like to suggest still, is to keep the old behavior of showing (for instance with 2 decimals) 9.18\times10^{-3} notation when exact p-values are requested, since that's actually exact, rather than rounding it to 2 decimals.