Paul,

Here are a couple of passages that I have found cause problems:

In particular, when an analysis uses methods to address missing data such as 
regression imputation, maximum likelihood, or non-response weights, the review 
process described in the last subsection of Section C (Analyses with Missing 
Data) should be followed instead, which includes an assessment of potential 
bias from using imputed data instead of actual data.

When the outcome measure is imputed for some subjects in the analytic sample, 
in addition to (c) and (d), the following data are required: (e) the means of 
the outcome measure for the subjects in the analytic sample with observed 
outcome data, separately for the intervention and comparison groups; (f) the 
means of the outcome measure for the subjects in the analytic sample with 
observed baseline and outcome data, separately for the intervention and 
comparison groups; (g) the standard deviations of the outcome measure for 
either the sample of subjects in the analytic sample with observed outcome data 
or the sample with observed baseline and outcome data; and (h) the number of 
subjects with observed outcome data in the analytic sample by condition. If 
these data are not reported in the study, the WWC will request them from the 
authors.

In practice, this has been interpreted as requiring a complete-case analysis in 
addition to the researcher's preferred approach. This allows WWC reviewers to 
essentially override the researcher's approach when they review the study.

I worked on a training program trial that used local college records for 
outcome measurement. The outcome Y was defined =1 if the local records 
indicated credential receipt and =0 otherwise. It was known that a subset of 
the sample attended some other college and that the subset was considerable 
larger in the control sample than in the treatment sample. So, we imputed 
credential receipt for people who attended other colleges. In this case, a 
complete case analysis would, I trust you agree, be strongly biased.  I was 
unwilling to prepare impact estimates that treated everyone not attending the 
local college as failing to have earned a credential. I failed to see how such 
alternate impact estimates could have been relevant to determining the quality 
of our impact estimates.

--Dave


From: Impute -- Imputations in Data Analysis 
[mailto:[email protected]] On Behalf Of Allison, Paul D
Sent: Wednesday, September 2, 2020 7:45 AM
To: [email protected]
Subject: Re: WWC, Imputation, and JSM

Warning from Abt: External email. Be careful opening links and attachments.

Hi Dave:

I'm not familiar with the WWC, but your email prompted me to take a look at 
their standards handbook at

https://urldefense.com/v3/__https://ies.ed.gov/ncee/wwc/Docs/referenceresources/wwc_standards_handbook_v4.pdf__;!!Dq0X2DkFhyF93HkjWTBQKhk!FK42zAwx6nEiF-EG441HwbntI195rgGIYVJYLfiwuz1CXoX1-WS8M9snyXJ1Yc9geltSho0B8BupBQ$
 
<https://urldefense.com/v3/__https:/ies.ed.gov/ncee/wwc/Docs/referenceresources/wwc_standards_handbook_v4.pdf__;!!Dq0X2DkFhyF93HkjWTBQKhk!AckH5BhOYbPDmObE0ynScTETKdcEKvZFJF8XFlC6ZoE8vD1nraio173bGltqUL4cSMMuCxA1DROD4Q$>


The section on missing data seems to take a more balanced view than your email 
suggests.  Am I missing something?



Paul




Paul D. Allison, Professor Emeritus
Department of Sociology
University of Pennsylvania
362 McNeil Building
3718 Locust Walk
Philadelphia, PA 19104-6299
215-898-6717

________________________________
From: Impute -- Imputations in Data Analysis 
<[email protected]<mailto:[email protected]>>
 on behalf of David Judkins 
<[email protected]<mailto:[email protected]>>
Sent: Tuesday, September 1, 2020 2:50 PM
To: 
[email protected]<mailto:[email protected]> 
<[email protected]<mailto:[email protected]>>
Subject: WWC, Imputation, and JSM


Dear ListServe Members,



Many of you probably know this, but the What Works Clearinghouse (WWC) in the 
United States has a strongly anti-imputation set of standards.  I have been 
told my studies will get failing grades unless I redo the analyses without 
imputed data.  I am trying to pull together a draft invited paper/panel session 
for JSM 2021 on this topic.  Anyone interested?



--Dave Judkins

Abt Associates









________________________________
This message may contain privileged and confidential information intended 
solely for the addressee. Please do not read, disseminate or copy it unless you 
are the intended recipient. If this message has been received in error, we 
kindly ask that you notify the sender immediately by return email and delete 
all copies of the message from your system.


________________________________
This message may contain privileged and confidential information intended 
solely for the addressee. Please do not read, disseminate or copy it unless you 
are the intended recipient. If this message has been received in error, we 
kindly ask that you notify the sender immediately by return email and delete 
all copies of the message from your system.

Reply via email to