Guidance about best practices for Missing Data at Item Level vs Variable Level

[Ranjana Dutta]

Hi, 

This may seem like trivial question to many, but I am looking for guidance on acceptable and suggested "best practices" for handling missing data at "item level" and at "variable level". 

I have a few multi-item measures (self-efficacy, personality, etc.) in addition to single indicator measures (e.g. GPA, years in college) administered to college students in an online survey. Often students miss a couple of items here and there within measures. At the present time, I am not doing item level analyses (e.g. Factor Analyses) of the multi-item measures. 

The total percentage missing for any item is not more than 2-5%. Case-wise, there are no more than 1% who have more than 50% data missing for any measure (or factor of a multidimensional measure).

My first questions is what is acceptable/best practice to compute the total-value on a scale with multiple items? (with that low numbers of missing values).

My second follow-up question is: Do I first deal with missing values at item level and then input variables in SPSS to see patterns of missingness in the scaled scores? (I am not familiar with Mplus, and do not have access to it.)

I would appreciate if anyone has an example published paper where this has been successfully done and can be modeled, or a resource which has some "how to" guidelines.

Thanks you in advance for any pointers you can provide.

[John F Hall]

Ranjana

 

You might find something in this which helps:

 

http://search.atomz.com/search/?sp-q=imputation&sp-a=sp1003afea&sp-advanced=1&sp-p=any&sp-w-control=1&sp-w=alike&sp-d=custom&sp-date-range=-1&sp-x=any&sp-c=25&sp-m=1&sp-s=0&sp-f=ISO-8859-1

 

 

John F Hall

 

[hidden email]

www.surveyresearch.weebly.com

 

 

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Ranjana Dutta
Sent: 30 September 2011 15:49
To: [hidden email]
Subject: Guidance about best practices for Missing Data at Item Level vs Variable Level

 

Hi, 

 

This may seem like trivial question to many, but I am looking for guidance on acceptable and suggested "best practices" for handling missing data at "item level" and at "variable level". 

 

I have a few multi-item measures (self-efficacy, personality, etc.) in addition to single indicator measures (e.g. GPA, years in college) administered to college students in an online survey. Often students miss a couple of items here and there within measures. At the present time, I am not doing item level analyses (e.g. Factor Analyses) of the multi-item measures. 

 

The total percentage missing for any item is not more than 2-5%. Case-wise, there are no more than 1% who have more than 50% data missing for any measure (or factor of a multidimensional measure).

 

My first questions is what is acceptable/best practice to compute the total-value on a scale with multiple items? (with that low numbers of missing values).

 

My second follow-up question is: Do I first deal with missing values at item level and then input variables in SPSS to see patterns of missingness in the scaled scores? (I am not familiar with Mplus, and do not have access to it.)

 

I would appreciate if anyone has an example published paper where this has been successfully done and can be modeled, or a resource which has some "how to" guidelines.

 

Thanks you in advance for any pointers you can provide.