MVA

[Chelminski, Iwona]

Hi Group,

What would you suggest for this scenario?
Over a half of the sample (700 out of 1300) were given a scale with one missing item, so for that variable I have 700 missing values. Would it be appropriate to use SPSS13  MVA (missing value analyses) to impute those values?
The total number of the items in the scale is 16 and there are no missing values for the other items.

If MVA is the way to go, which method would you use, EM or Regression?

Thanks in advance,

Iwona

[Melissa Ives]

Iwona,

First, I haven't used MVA, so I don't have an answer for that, but in
terms of scale creation...

If you have a true scale (reliability (Cronbach's alpha > .7) and only
one missing item AND the missing-ness of that item is not systematic
(e.g. some particular group was not asked the item or the scale changed
half way through, etc.), then you should be able to impute the missing
value in the calculation of the scale.

We use a formula like this (assuming a 7 item scale)

        compute scalename=rnd(mean.3(scalevariablelist)*7).

The rnd will return the scale to the original metric (no decimals)
The mean.3 means that there must be at least 3 valid items on the scale
to create it.
The *7 multiplies the mean by the number of items--so you end up with an
additive scale AS IF it had responses for all valid items (in essence,
this is a mean replacement for missing items).  The 7 is a number that
represents the number of items in the scale and will change accordingly.

If you don't have a true scale OR do have a systematic reason for
missing data, this is not the best option.  We generally then impute the
values of the new items with regression (predict the scale from the
other items).

Melissa

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Chelminski, Iwona
Sent: Wednesday, December 20, 2006 10:15 AM
To: [hidden email]
Subject: [SPSSX-L] MVA

Hi Group,

What would you suggest for this scenario?
Over a half of the sample (700 out of 1300) were given a scale with one
missing item, so for that variable I have 700 missing values. Would it
be appropriate to use SPSS13  MVA (missing value analyses) to impute
those values?
The total number of the items in the scale is 16 and there are no
missing values for the other items.

If MVA is the way to go, which method would you use, EM or Regression?

Thanks in advance,

Iwona



PRIVILEGED AND CONFIDENTIAL INFORMATION
This transmittal and any attachments may contain PRIVILEGED AND
CONFIDENTIAL information and is intended only for the use of the
addressee. If you are not the designated recipient, or an employee
or agent authorized to deliver such transmittals to the designated
recipient, you are hereby notified that any dissemination,
copying or publication of this transmittal is strictly prohibited. If
you have received this transmittal in error, please notify us
immediately by replying to the sender and delete this copy from your
system. You may also call us at [phone] for assistance.

[statisticsdoc]

Stephen Brand
www.statisticsdoc.com

Iwona,

I take it that half of the sample were presented with a copy of the survey that was missing the item in question.  If responses on the missing item, or the underlying construct, differed between the subjects who received the complete and the incomplete version of the scale, then I would be concerned that data on the item is not missing at random.  The procedures for imputing missing data in SPSS assume that the data are missing at random.  I would suggest that you compare subjects who received the full and the incomplete version of the survey on their mean score on the 15 items that they have in common.

The fact that the data on the item are missing for about half of the cases is also a concern.  There is no hard and fast rule about how much complete data is needed for analysis.  Suggested cutoffs for the maximum allowable missing data range from 20% to 50%, with the caveat that with higher proportions of missing data, the model for imputing missing values has to have very good fit.

Instead of imputing missing data, a conservative course of action is to compute the scale scores using the 15 items what were presented to all of the participants.  You could analyze the data for the portion of the sample that received the full scale to see whether "Item 16" makes an important contribution to the internal consistency of the scale (what is alpha with the item deleted?)  I would also consider whether the item makes an important contribution to the scope and validity of the scale.  Frankly, I would tend to take this course of action in this situation, given the scope and structure of the missing data.

If you decide to substitute the missing values, most references  recommend against pro-rating the scale with the subjects' mean response from the 15 complete items.  The subjects' mean of the 15 completed items may not provide the most accurate estimate of their response to the missing 16th item.  Nor would it be a good idea to substitute the mean of the missing item; this substitution can also reduce the variance of the scale and its correlation with other variables.

Carrying out an MVA and imputing missing values are preferable to pro-rating scores or using mean substitution.  If you decide to go this route, then the EM method is may be preferable to Regression because it makes fewer assumptions about the data.

I hope this is of some help.

Best,

Stephen Brand

--
For personalized and experienced consulting in statistics and research design, visit www.statisticsdoc.com

Iwona Chelminski Wrote:

Hi Group,

What would you suggest for this scenario?
Over a half of the sample (700 out of 1300) were given a scale with one missing item, so for that variable I have 700 missing values. Would it be appropriate to use SPSS13  MVA (missing value analyses) to impute those values?
The total number of the items in the scale is 16 and there are no missing values for the other items.

If MVA is the way to go, which method would you use, EM or Regression?

Thanks in advance,

Iwona