Date: Sun, 20 Apr 2003 08:29:22 -0400
Reply-To: Art@DrKendall.org
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Arthur J. Kendall" <Art@DrKendall.org>
Organization: Social Research Consultants
Subject: Re: Argument for keeping outliers: urgent help needed
Content-Type: text/plain; charset=us-ascii; format=flowed
The argument should be the other way around. One would need to have
very strong reasons to exclude values for some cases.
It is very hard to believe there would be outliers on a scale. If the
response scale is wrong/right, for 10 items and the score says that one
person had 12 wrong that would be a suspected outlier. If there are 10
items, with a 1 to 5 response scale, then scores below 10 or over 50 are
suspected outliers.
Based on consulting on stat and methodology for over 30 years, I believe
the first thing to suspect when there are suspicious values is failure
of the quality assurance procedure. In my experience, rechecking qa
typically eliminates over 80% of suspicious data values.
How many items were there in your scale?
What is the response scale?
What makes someone suspect outliers?
What are the z-scores of the suspicious values both when the they are
included in the mean and sd and when the mean and sd are based on the
other cases?
Was this a paper-and-pencil instrument? Do you have access to the paper
copies?
Were items from different scales intermixed?
Were items within the scale balanced?
Did you check the item direction keying?
Did you check the item inclusion keying?
Did you double enter your data or proofread the entered data?
Who were the respondents?
What was their motivation to participate?
Was the topic "boring" or sensitive?
If you look at the original instruments for individuals with suspicious
scores, is there clear evidence that they "pattern responded"?
alternating t/f or other systematic pattern answering without regard to
the stem?
Can you re-administer the instruments, at least to the individuals with
suspicious scores?
What does the frequency analysis look like?
How did the reliability analysis come out?
Art
Art@DrKendall.org
Social Research Consultants
University Park, MD USA
(301) 864-5570
A B wrote:
> Hello,
>
> I have a scale questionnaire and decided to keep the outliers since there is no evidence that the outliers are not from the intended population.
>
> I'm convincing with this decision. But I need to defend such a decision in my academic research. Please help me find the most convincing argument for that.
>
> Best regards,
>
>
>
>
>
> ---------------------------------
> Do you Yahoo!?
> The New Yahoo! Search - Faster. Easier. Bingo.
>
