Hey, the app shows “not enough data yet” for medication correlations, even though I have been logging it for more than 4 days (and also filled past few months historical usage manually)
Hi @Vvvv!
The three days minimum is only a minimum. At that point it’s possible for the maths to start working, but it doesn’t mean it will tell you anything real or useful.
You could try opening up the filters view (the sliders icon at top right) and sliding “Maximum correlation width” all the way to the right, to 1.9. Though if that shows some correlations that were previously filtered out, they’re not going to be meaningful yet. They’ll be more like random noise.
Correlations don’t typically start becoming statistically sensible until you’ve got about a month or more of data. And even then the results will mostly be unreliable noise. At maybe 2-3 months of data I think you should start to see correlations that might actually mean something.
Such is the nature of statistics! Anyway, I expect you should start to see some correlations appearing once you’ve got more than a week or two of data. Though hold off in reading too much into them until you’ve go a month or more of data.
Thank you for reply. I have entered historical data for the past 97 days (medication is taken daily), but I still see the same message. Is it possible more data is needed?
Also only a few entries of other metrics, like daily water seem to work, while medications show no correlations
hm. 97 days should be enough! It’s possible it’s still updating the correlations. Or also possible that there’s not enough overlapping data.
Do you own an Apple Watch or anything else that records heart rate data etc? Actually even with just an iPhone and no wearable recording throughout the day, there should still be data types with data for every day. The iPhone creates various daily data points too.
But yeah, if you’ve got data from a wearable then there’ll be much more there to build correlations from. So with 97 days of medication data I’d expect to definitely already see some correlations for the medication data types showing up.
Maybe worth waiting a minute on the Correlations List view, to see if it fills in soon after. It might be still trying to build the initial correlations, which can take some moments.
I have quite a lot of data from Apple Watch, and all other categories do show correlations. I waited for a few minutes with the medication view opened but it still showed “not enough data yet”. Also with other categories, phone gets a little warm, while with medications open it does not so i guess no calculations are happening. Also left correlations list view opened for some time but it did not work either
Hm. That’s curious. It sounds like you should have enough data now - easily enough. But yeah if it seems like it’s already finished building all the correlations for that list view and nothing’s showing… it’s decided none of them meet the quality thresholds.
What happens if you slide that “Maximum confidence width” slider in the filters view all the way to the right, to 1.9?
Nothing changes by adjusting sliders 
also I just noticed that this happens only with one type of medication, the one which has the most entries (and which I am most interested in). With other medications it works fine.
Could this be related to the fact that I entered historical data after I first opened the app? Although I tried to reinstall it but it didn’t work
That won’t be it. What you did there is fine.
Hmm. I wonder if there’s something about that medication’s values that mean it’s producing useless correlations. Medication is a tricky one, because for most of our medications we tend to take the same amount every day. So really the only thing the correlation maths can pick up is days with / days without, ie 0 versus 1. It means the maths doesn’t maths very well.
Though even with that said, you should still be already seeing correlations, even if they’re not particularly useful ones.
I wonder if it might be the lack of zero values, ie days without the medication. I’m not sure how the current code deals with that for medications. I think if you mark a dose as “skipped” that’ll log as a zero value, which is useful information for the maths. But I’ll double check the code.
I’m also going to add a new feature in the next update, for “zero filling” days. Like, if the data only ever carries information for days when something was present, and never data for its absence, that somewhat ruins the maths. But that’s kind of how a lot of data is in the Health app’s database.
So the new feature will basically be a calendar view, showing you all the days you do have data for a data type. And then you can select missing days and choose to “zero fill” those, to add explicit “zero dose” equivalent to each of those. That will improve the maths and correlations significantly.
The reason why This & That doesn’t automatically treat all “no data” days as “zero” is because there’s a big difference between no information and a confirmed absence. Like, what if you forget to log something for a whole month? Should that all be treated as “didn’t take medication”? That’d mess up the maths even worse.
Anyway yeah, that new feature should help a lot in this situation. Though I’m still unsure why you’re not seeing correlations for this specific medication yet! That is quite annoying, I’m sure.
I’ll have a poke around the code and see if I can find any more clues… Maybe there’s some more conditions for why it might be filtering out low quality correlations. Though that would still only tell us why it’s filtering them, not how to improve the quality of them. Which unfortunately can only be done by having more data, with the maths improving in quality over time.
The other medication also don’t have any skipped values but they do pick up correlations. This one was taken every day for 100 days at simmilar time of the day
Let me know if you need help from my side too to help debug this. IMO this is a very interesting app with lots of potential. If not for this bug I would definitely buy full time subscription 
1 Like
Thanks!
I’ve also got a bunch of medications I’m wanting to get better correlation results for. So far only one of my medications is showing any useful correlations. And for that one it’s because it’s a different amount each day, depending on how many doses I take. So that creates more of a signal for the maths to pick up on. For my other medications the correlations are there but none of them are of any use / make any sense yet.
Medications is actually why I first made this app! I wanted to see what benefits they were having. But after getting started on building it I realised that Apple didn’t at that time make the medication data available to third party apps 
Though they did finally add it. But that means the medications system in This & That is one of the newest parts, and … yeah, I think it still needs some improvements.
The “zero filling” feature will probably be the biggest help, so that’s next on my todos. If I get some time before the end of the year (I’m taking a couple of weeks off in Jan) I’m hoping to get an update out with that new feature!
Is it possible days where medication was not taken are not counted towards correlations? I only started getting correlations when the dose was changed and now it only shows correlation related to the dose change. Basically 0→1mg no correlations (not enough data), 1mg→2mg lots of correlations (and charts only show days when 1mg and 2 mg of medication was taken, ignoring days before medication was started)
@Vvvv Yeah very much that! If there’s nothing logged for a day, there’s no data for the calculation. Though if you log a dose as “skipped” that should result in a 0mg data point for the day.
This is where that “zero filling” feature will come in handy. I need to get onto building that one soon!