I have a question about which core places version we should use for the “Weekly Places Patterns Backfill for Dec 2020 and Onward Release”?

Questions & Help
, ,
21

1 - no but i can ask them to do this 2 - I asked them to look into this as well but haven’t heard back 3 - yes.

22

thanks for looking into it!

23

let me know if you can’t replicate the result; it’s possible the student made an error.

24

Great, this is helpful. It would be awesome if they could check for a couple of other random weeks! And if there are any patterns about the missing POIs, please let me know

25

Also, which column was used to join the datasets?

26

safegraph_place_id

27

she seems to think it’s pretty random

28

or at least, she doesn’t notice any patterns

29

i think it happens in other weeks as well

30

i took a look myself and confirmed that this is an issue; there are definitely rows in patterns missing from core places 2020 - 11

31

there also does seem to be some pattern in the rows which are missing. for example

32

i think they tend to have fewer visits, and they also seem to be a different set of brands

33

among POIs that ARE missing from core places, most common place names are stuff like “Metro by T Mobile Authorized Dealer” and “Suzuki”

34

among POIs that are not missing from core places, most common place names are things like

35

“USPS” and “Subway” and “Dollar General” and “Shell Oil” among the non-missing data.

i am guessing that what’s happening here is something like…brands are going out of business or the stores are vanishing somehow…not sure…and so they’re not ending up in core places 2020 - 11. not sure.

36

and, also, at least in the week i’m looking at, a ton of them seem to be churches (among the missing rows)

37

so the data’s definitely not missing at random. any insight you can lend is very helpful.

38

@Emma_Pierson_Stanford Thanks! This will be very helpful to me! I hope to have some answers for you today

39

thank you very much!

40

Looking at Weekly Patterns 2019-01-07, I was able to recreate the issue with about 4.1% of POIs missing from Nov 2020 Places.

There was 100% coverage when I used Dec 2020 Places. I’m checking with the team to see what’s going on here—it seems like something was mislabeled or left ambiguous. I’ll let you know when it’s resolved.

For your analysis, you can move forward with the Dec 2020 Places. I would expect other weeks to have full coverage in Dec 2020 Places as well. If you find that’s not the case, let me know!