Preprocessing Issues

[palday]

We've found a small pre-processing issues:

• [x] subid isn't unique across labs, so currently the model treats them as partially crossed instead of fully nested. If subjects didn't move between labs, then this is easy to fix: d_lmer$subid <- interaction(d_lmer$subid, d_lmer$lab)
• [x] Some of the CSVs encode NA as N/A. This can be fixed by changing your read_csv() calls to read_csv(...,na=c("","NA","N/A")).
• [x] There are a number of parsing errors when reading 02_validated_output.csv.
• [x] At some point in the manuscript, the number of labs following exclusion is dropped from 69 to 67. But there are still 69 labs in the dataset going into the model (based on running all the code up to the fitting of mod_lmer). This isn't to do with missing levels of a factor either: length(unique(d_lmer$lab)) still yields 69. Perhaps this emerges from the parsing errors and things not dropping/merging/etc. like this should.
• [x] Non monolingual babies are in the monolingual data set.

To the extent that we've been able to work around these, none of these seem to have a real impact on the results, but the nested vs. partially crossed issue for subid can have a big impact on compute time.

[dmbates]

Perhaps it will help to have a list of the subid values that are repeated in different labs

"1", "10", "11", "1101", "1104", "1106", "1110", "1111", "1115", "12", "13", "14", "15", "16", "17", "18", "19", "2", "20", "21", "22", "23", "24", "25", "26", "27", "28", "3", "30", "31", "32", "33", "34", "37", "38", "39", "4", "41", "42", "4204", "45", "46", "5", "56", "6", "64", "7", "71", "75", "76", "8", "80", "9", "mb001", "mb002", "mb003", "mb004", "mb005", "mb006", "mb007", "mb008", "mb009", "mb01", "mb010", "mb011", "mb012", "mb013", "mb014", "mb015", "mb016", "mb017", "mb018", "mb019", "mb01b", "mb02", "mb020", "mb021", "mb022", "mb023", "mb024", "mb025", "mb026", "mb027", "mb028", "mb029", "mb03", "mb030", "mb031", "mb032", "mb033", "mb034", "mb035", "mb036", "mb037", "mb038", "mb039", "mb04", "mb040", "mb041", "mb042", "mb043", "mb044", "mb045", "mb046", "mb047", "mb048", "mb05", "mb06", "mb065", "mb066", "mb069", "mb07", "mb070", "mb071", "mb072", "mb073", "mb074", "mb075", "mb076", "mb08", "mb09", "mb0903", "mb1", "mb10", "mb1002", "mb102", "mb103", "mb104", "mb105", "mb107", "mb108", "mb11", "mb1103", "mb12", "mb1201", "mb1202", "mb1203", "mb1204", "mb1205", "mb1206", "mb1207", "mb1208", "mb13", "mb14", "mb15", "mb16", "mb17", "mb18", "mb19", "mb2", "mb20", "mb21", "mb22", "mb23", "mb24", "mb25", "mb26", "mb27", "mb28", "mb29", "mb3", "mb30", "mb31", "mb32", "mb33", "mb34", "mb35", "mb36", "mb37", "mb38", "mb39", "mb4", "mb40", "mb41", "mb42", "mb43", "mb45", "mb46", "mb48", "mb49", "mb5", "mb50", "mb6", "mb7", "mb8", "mb9", "p01", "p02", "s01", "s02", "s03", "s04", "s05", "s06", "s07", "s08", "s09", "s10", "s11", "s12", "s13", "s14", "s15", "s16", "s17", "s18", "s19", "s20", "s21", "s22", "s23", "s25", "s26", "s27", "s28", "s29", "s30", "s31", "s32", "s33", "s34"

For some reason "mb10" is the most popular subject label, used in 13 labs "babylabbrookes" "babylabkonstanz" "baldwinlabuoregon" "cdcceu" "childlabmanchester" "escompicbsleipzig" "kokuhamburg" "lcduleeds" "madlabucsd" "minddevlabbicocca" "musdevutm" "pocdnorthwestern" "weescienceedinburgh"

`

[dmbates]

I think I saw in the manuscript that each lab used only one method but there are five labs that use multiple methods

│ 1 │ babylabpotsdam │ hpp │ │ 2 │ babylabpotsdam │ singlescreen │ │ 3 │ infantcogubc │ singlescreen │ │ 4 │ infantcogubc │ eyetracking │ │ 5 │ lancaster │ singlescreen │ │ 6 │ lancaster │ eyetracking │ │ 7 │ ldlottawa │ singlescreen │ │ 8 │ ldlottawa │ eyetracking │ │ 9 │ nusinfantlanguagecentre │ eyetracking │ │ 10 │ nusinfantlanguagecentre │ singlescreen │

[mcfrank]

Thanks for these extremely helpful comments! I do think the ms acknowledges the multi-method labs (e.g., in the table of participating labs) but I will check if there is a presupposition of choice.

@philip great catches, I will work on these when we next take up the revisions!

On Wed, May 22, 2019 at 1:16 PM Douglas Bates notifications@github.com wrote:

I think I saw in the manuscript that each lab used only one method but there are five labs that use multiple methods

│ 1 │ babylabpotsdam │ hpp │ │ 2 │ babylabpotsdam │ singlescreen │ │ 3 │ infantcogubc │ singlescreen │ │ 4 │ infantcogubc │ eyetracking │ │ 5 │ lancaster │ singlescreen │ │ 6 │ lancaster │ eyetracking │ │ 7 │ ldlottawa │ singlescreen │ │ 8 │ ldlottawa │ eyetracking │ │ 9 │ nusinfantlanguagecentre │ eyetracking │ │ 10 │ nusinfantlanguagecentre │ singlescreen │

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/manybabies/mb1-analysis-public/issues/5?email_source=notifications&email_token=AAI25F5GCKXDDNCUB7PVJT3PWWSYDA5CNFSM4HORLC22YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWAGPII#issuecomment-494954401, or mute the thread https://github.com/notifications/unsubscribe-auth/AAI25F747SRVV3KLACLSS73PWWSYDANCNFSM4HORLC2Q .

[christinabergmann]

Re: monolingual some more info (after having to remember myself) Labs might differ in their definition, so we re-calculated monolingual status and added the column monolingual (so there might still be participants where it says bilingual in lang_group).

[mcfrank]

thanks, have addressed these.

• subid_unique used for lmers
• read csv issue fixed, thanks
• parsing errors are dealt with once validation happens
• I believe this mismatch was due to the child="exclusions.Rmd" issue in #8
• as @christinabergmann says, this is a categorization issue dealt with in variable validation