Low variability issues

[Mark Webb-5]

I have a set of agreement statements designed to measure staff engagement. The data has very low variability between variables (statements) and within respondents. All statements very negatively skewed.  Each statement far from normally distributed. Most claims to be happy with all statements. If i run a factor analysis there is basically one factor with ev gt 1. CFA in Rs lavaan also struggles. I tried a log10 transformation to help the normality issue with little improvement.

I think the underlying issue is that the statements are not "contentious" enough.

What tests are available to analyse non-normal, low variability data?

Or is the only solution to measure issues in such a way as to get more variability and more normal data distributions?

Any comments on "boring" low variability data and poor normality welcome.

Thank you in advance.

Mark

--
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[John F Hall]

How many items are there, and how many respondents? 

Have they almost all agreed/disagreed with every (uni-polar) statement. 

Has one or more persons filled out the battery several times? 

Looks like you need more contentious items or, perhaps better, a set of bi-polar choices (in semantic differential format).

 

John F Hall  MA (Cantab) Dip Ed (Dunelm)

[Retired academic survey researcher]

 

Email:          [hidden email]

Website:     Journeys in Survey Research

Course:       Survey Analysis Workshop (SPSS)

Research:   Subjective Social Indicators (Quality of Life)

 

From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Mark Webb
Sent: 20 February 2019 14:56
To: [hidden email]
Subject: Low variability issues

 

I have a set of agreement statements designed to measure staff engagement. The data has very low variability between variables (statements) and within respondents. All statements very negatively skewed.  Each statement far from normally distributed. Most claims to be happy with all statements. If i run a factor analysis there is basically one factor with ev gt 1. CFA in Rs lavaan also struggles. I tried a log10 transformation to help the normality issue with little improvement.

I think the underlying issue is that the statements are not "contentious" enough.

What tests are available to analyse non-normal, low variability data?

Or is the only solution to measure issues in such a way as to get more variability and more normal data distributions?

Any comments on "boring" low variability data and poor normality welcome.

Thank you in advance.

Mark

 

--

Mark Webb [phone] [phone]

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Rich Ulrich]

Well, what were you hoping to find out?

Apparently, "most" are very happy with "almost everything".

Are there cracks in that happy facade? That is, are SOME items

outliers?  Are a few people outliers?

Given the universal role of "response bias", this could be a case where

dichotomizing the items is justifiable -- that is, there might not be

much meaning in the distribution outside of not-fine.  You can find

general outliers by tabulating for each person how many times

they were not at the extreme.

Analyses should be guided by what you want to say. Transformations

can be convenient for simplifying the descriptions, which, here, is

probably more important than for "selecting the test." Get a good

description of < whatever you want to say > ; and then, consider

whether any testing is useful.  I did not notice any hypotheses

mentioned in your post.

--

Rich Ulrich

---

From: SPSSX(r) Discussion <[hidden email]> on behalf of Mark Webb <[hidden email]>
Sent: Wednesday, February 20, 2019 8:55 AM
To: [hidden email]
Subject: Low variability issues

 

I have a set of agreement statements designed to measure staff engagement. The data has very low variability between variables (statements) and within respondents. All statements very negatively skewed.  Each statement far from normally distributed. Most claims to be happy with all statements. If i run a factor analysis there is basically one factor with ev gt 1. CFA in Rs lavaan also struggles. I tried a log10 transformation to help the normality issue with little improvement.

I think the underlying issue is that the statements are not "contentious" enough.

What tests are available to analyse non-normal, low variability data?

Or is the only solution to measure issues in such a way as to get more variability and more normal data distributions?

Any comments on "boring" low variability data and poor normality welcome.

Thank you in advance.

Mark

--
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD