missing values imputation analysis

[Paola Chivers-2]

Hi,

 

I am creating an imputed file (20 imputed datasets).  I want to test whether the imputed values are significantly different from the original non-imputed data.

 

I guess I can do this two ways (or both)

1.       compare the original data set to the pooled data set, or

2.       compare the non-missing values to the missing imputed values.

 

I’m not quite sure how to get SPSS to do this for me.

 

Under the missing values imputation I can ask for descriptives (/IMPUTATIONSUMMARIES MODELS DESCRIPTIVES).  As my imputation takes over a day, it is not a simple exercise to test this and see.

 

Can anyone tell me how to do this in spss, and whether in fact the descriptives option in the imputation will provide this information.  Any other information for testing my imputed data or recommended reading would also be appreciated.

 

Thanks.

 

Regards,

Paola

 

“Ours has become a time-poor society, fatigued by non-physical demands and trying to compartmentalize daily living tasks.  It is small wonder that physical activity is discarded in this environment” p126 (Steinbeck, 2001)

 

P Please consider the environment before printing this email.

 

 

[Paola Chivers-2]

Hi,

 

I apologise in advance for all these queries ... but I have limited access and time to the missing values module.

 

Thanks Jason for your information.  After a 5+ hours run to impute a dataset I have confirmed what Jason described (and is described in the case study). 

 

However how do I statistically compare whether there is a statistically significant difference between my original dataset, individual imputed data sets, and the pooled data set?  I guess I’m trying to do a t-test but on the imputed data sets? 

 

I have run an Independent-samples t-test with the imputation variable as my grouping variable.  I then only can assign two groups.  In one case I ran 0 (original dataset) and 1 (first imputed dataset), in another run 0 and 20.  The output always gives me the dataset I requested and pooled results – which are always the same.  Should I be running an ANOVA which compares all the groups?

 

Secondly, with my t-test results there are some variables where there is a significant difference between my original data and my pooled imputed data result!  What could be causing this?  Could it be skewness or kurtosis?  If so, how does transformation affect imputation and further analysis?

 

If I ever get my head around these issues J, how best should I use this imputed file in AMOS (I don’t think it uses a pooled variable – but that’s because my reading hasn’t uncovered this or how yet).

 

Regards,

Paola

 

“Ours has become a time-poor society, fatigued by non-physical demands and trying to compartmentalize daily living tasks.  It is small wonder that physical activity is discarded in this environment” p126 (Steinbeck, 2001)

 

P Please consider the environment before printing this email.

 

 

From: [hidden email] [mailto:[hidden email]]
Sent: Monday, 8 June 2009 10:45 AM
To: Paola Chivers
Subject: Re: missing values imputation analysis

 

What you request is a by product of the procedure. I found the case study (Help>Case Study...) a great place to start.

Yes.

Descriptive statistics tables show summaries for variables with imputed values. A separate table is produced for each variable. The types of statistics shown depend on the measurement level of the variable imputed. The descriptive statistics will provide you three aspects, by default, including:

a. summary of the original dataset
b. summary of imputed values split for each imputation
c. summary of each imputed dataset (including non-missing and imuted values)

Cheers,


Jason

On 08/06/2009 11:52am, Paola Chivers <[hidden email]> wrote:
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> Hi,
>
>
>
>
>  
>
>
>
>
> I am creating an imputed file (20 imputed datasets).  I
>
> want to test whether the imputed values are significantly different from the
>
> original non-imputed data.
>
>
>
>
>  
>
>
>
>
> I guess I can do this two ways (or both)
>
>
>
> 1.       compare
>
> the original data set to the pooled data set, or
>
>
>
> 2.       compare
>
> the non-missing values to the missing imputed values.
>
>
>
>
>  
>
>
>
>
> I’m not quite sure how to get SPSS to do this for me.
>
>
>
>
>  
>
>
>
>
> Under the missing values imputation I can ask for
>
> descriptives (/IMPUTATIONSUMMARIES MODELS DESCRIPTIVES).  As my imputation
>
> takes over a day, it is not a simple exercise to test this and see.
>
>
>
>
>  
>
>
>
>
> Can anyone tell me how to do this in spss, and whether in
>
> fact the descriptives option in the imputation will provide this information. 
>
> Any other information for testing my imputed data or recommended reading would
>
> also be appreciated.
>
>
>
>
>  
>
>
>
>
> Thanks.
>
>
>
>
>  
>
>
>
>
> Regards,
>
>
>
>
> Paola
>
>
>
>
>  
>
>
>
>
> “Ours has become a time-poor society, fatigued by
>
> non-physical demands and trying to compartmentalize daily living tasks. 
>
> It is small wonder that physical activity is discarded in this
>
> environment” p126 (Steinbeck, 2001)
>
>
>
>
>  
>
>
>
>
> P Please consider the
>
> environment before printing this email.
>
>
>
>
>  
>
>
>
>
>  
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>