Coding Missing quantitative variable in 3dLMEr

andrewlynnphd · February 21, 2024, 9:27pm

I'm running into an issue where I have missing data for a quantitative variable in my 3dLMEr model. Using a blank space (' ') to hold the spot where the data should be listed results in the following error. Is there a way to code missing quantitative variables?

 ** ERROR: Duplicates found in "InputFile" column!!!
-- Note: Listed files are AFTER the 1st instance of the duplicates.
[66]   DDK2007
[68]   5
[69]   0.030784
[156]  digit
[157]  5
[160]  DDK2018
[161]  dot
[162]  5
[163]  0.30382
[166]  digit
[167]  5
[170]  DDK2019
[171]  dot
[172]  5
[173]  0.62199
[224]  digit
[225]  5
[228]  DDK2028
[229]  dot
[230]  5
[231]  0.18441
[258]  digit
[259]  5
[262]  DDK2031
[263]  dot
[264]  5
[265]  0.10069

In the code below you'll see that Subj DDK2028 is missing the "mathtalk" variable. I've also tried replacing missing data with "NaN" but that appears to result in other issues.

AFNI version info (afni -ver): Precompiled binary macos_13_ARM_clang: Jan 24 2024 (Version AFNI_24.0.02 'Caracalla')

3dLMEr -prefix 3dLMEr_output_Format_Quantity_HomeMath_linear+tlrc \
-jobs 12 -mask /Users/andrewlynn/Documents/Projects/HomeMath_RSA/ref/mask/BN_Atlas_HaskinsPeds_NL_template_2.5_mask+tlrc.HEAD \
-model "format*quantity*mathtalk+motion+(1+mathtalk|Subj)+(1+mathtalk|format:Subj)+(1+mathtalk|quantity:Subj)" \
-qVars "mathtalk,motion" \
-qVarCenters "2.85,0.16" \
-SS_type 3 \
-dataTable \
Subj    format  quantity    mathtalk    motion  InputFile \
DDK2001 digit 2 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Digit_2#0_Coef]" \
DDK2001 digit 3 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Digit_3#0_Coef]" \
DDK2001 digit 4 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Digit_4#0_Coef]" \
DDK2001 digit 5 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Digit_5#0_Coef]" \
DDK2001 digit 6 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Digit_6#0_Coef]" \
DDK2001 digit 7 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Digit_7#0_Coef]" \
DDK2001 dot 2 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Dots_2#0_Coef]" \
DDK2001 dot 3 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Dots_3#0_Coef]" \
DDK2001 dot 4 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Dots_4#0_Coef]" \
DDK2001 dot 5 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Dots_5#0_Coef]" \
DDK2001 dot 6 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Dots_6#0_Coef]" \
DDK2001 dot 7 2.4583 0.14815 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2001/Y1/proc/exemplar.stats.DDK2001+tlrc"[Dots_7#0_Coef]" \
...
DDK2028 digit 2   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Digit_2#0_Coef]" \
DDK2028 digit 3   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Digit_3#0_Coef]" \
DDK2028 digit 4   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Digit_4#0_Coef]" \
DDK2028 digit 5   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Digit_5#0_Coef]" \
DDK2028 digit 6   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Digit_6#0_Coef]" \
DDK2028 digit 7   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Digit_7#0_Coef]" \
DDK2028 dot 2   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Dots_2#0_Coef]" \
DDK2028 dot 3   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Dots_3#0_Coef]" \
DDK2028 dot 4   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Dots_4#0_Coef]" \
DDK2028 dot 5   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Dots_5#0_Coef]" \
DDK2028 dot 6   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Dots_6#0_Coef]" \
DDK2028 dot 7   0.18441 /Users/andrewlynn/Documents/Projects/HomeMath_RSA/Processed_Data/DDK2028/Y1/proc/exemplar.stats.DDK2028+tlrc"[Dots_7#0_Coef]" \
...

Gang · February 22, 2024, 3:55pm

No imputation methods are available in 3dLMEr. If your missing data is unrelated to task conditions, exclude rows with missing values from the data table.

Additionally, the specified model raises a potential concern:

-model "format*quantity*mathtalk+motion+(1+mathtalk|Subj)+(1+mathtalk|format:Subj)+(1+mathtalk|quantity:Subj)" \

If mathtalk remains constant across different format and quantity levels, consider:

-model "format*quantity+mathtalk+motion+(1|Subj)+(1|format:Subj)+(1|quantity:Subj)" \

Gang Chen

andrewlynnphd · April 16, 2024, 7:14pm

Would it be possible to obtain a binary voxel-wise map of model fits that are singular? This way if it's just a few voxels then one could move forward and ignore those voxels, but if it's the whole brain then one could update the model specs?

Gang · April 17, 2024, 3:46pm

I wouldn't worry too much about those "singular" warnings. They're usually harmless in this context, as they often occur when one or more variance estimates are close to zero.

Gang Chen