compound variables - kruskall wallis
Locked
 
|
Wanting to test a compound variable (i.e. combined 6 variables from a battery
of questions which ask about attitudes to gaming) against a variable about frequency of playing games (4 categories). Using kruskall wallis in SPSS but not sure it is legitimate to use compound variables in this test? Only Non parametric tests are possible with this data - does not meet any parametric conditions. All help appreciated. ===================== 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 |
Re: compound variables - kruskall wallis
|
|
Kruskal wallis will not solve your problem On 19/11/2013 15:43, "Oscar M" <odriscollfm@...> wrote: Wanting to test a compound variable (i.e. combined 6 variables from a battery Professor Diana Kornbrot email: : d.e.kornbrot@... web: http://dianakornbrot.wordpress.com/ http://go.herts.ac.uk/diana_kornbrot Work Department of Psychology School of Life and Medical Sciences University of Hertfordshire College Lane, Hatfield, Hertfordshire AL10 9AB, UK voice: +44 (0) 170 728 4626 Home 19 Elmhurst Avenue London N2 0LT, UK voice: +44 (0) 208 444 2081 mobile: +44 (0) 740 318 1612 |
|
|
Thanks Diana Can you or anyone else tell me why Kruskal Wallis is not OK in
this case? Thanks From: Kornbrot, Diana
[mailto:[hidden email]] You need ordinal regression also known as
plum in spss Wanting to test a compound variable (i.e.
combined 6 variables from a battery Professor
Diana Kornbrot |
Re: compound variables - kruskall wallis
|
In reply to this post by Oscar M
please explain what you
mean by a compound variable.
Is it a summative scale, i.e., a sum or mean of dichotomous or other interval level scales, e.g., with a Likert response scale? What was the response scale on the variable about frequency? Who wrote it? Art Kendall Social Research ConsultantsOn 11/19/2013 10:59 AM, Oscar M [via SPSSX Discussion] wrote: Wanting to test a compound variable (i.e. combined 6 variables from a battery
Art Kendall
Social Research Consultants |
Re: compound variables - kruskall wallis
Administrator
|
In addition to what Art has asked for:
1. Please clarify which variable is the explanatory variable and which the outcome (dependent) variable. (Diana assumed that the variable with 4 ordered frequency of playing categories was the outcome when she suggested PLUM. But your suggestion of Kruskal-Wallis makes me think you view the "compound variable", as you call it, the outcome.) 2. Please clarify what you mean when you said that only nonparametric tests are possible because the data do not meet "any parametric conditions". What conditions? Comment: Parametric tests are far more robust than many people realize, and rank-based nonparametric tests (when used to test null hypotheses about location) are far less robust than people realize. (You can find an article by Gene Glass & colleagues that talks about the 1950s "stampede" to nonparametric methods, and the fact that it was largely unnecessary.) So I'm quite curious about what you're seeing that invalidates parametric tests. HTH.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
Re: compound variables - kruskall wallis
|
|
In reply to this post by Oscar M
By compound I assume you mean that variable attitudes is sum of score on 6 items form your batteries? q1 is attitudes normally distributed? If so can use ANOVA with frequency category as categorical predictor. Chose polynomial in contrasts to check if there is linear rise with frequency category If not try PLUM read spss help carefully for interpretation of output. Or google for good internet source on ordinal regression Why not K-W? Because it assumes that, although not normal , dependent variable has same distributions shape for all 4 categories. This is rare as snowball in hell for summed scales. KW was aimed at variables like time and income that are continuous but long tailed and is widely misused for bounded scales such as your compound scale Best Diana On 19/11/2013 19:11, "Mike O'Driscoll" <odriscollfm@...> wrote: Thanks Diana Professor Diana Kornbrot email: : d.e.kornbrot@... web: http://dianakornbrot.wordpress.com/ http://go.herts.ac.uk/diana_kornbrot Work Department of Psychology School of Life and Medical Sciences University of Hertfordshire College Lane, Hatfield, Hertfordshire AL10 9AB, UK voice: +44 (0) 170 728 4626 Home 19 Elmhurst Avenue London N2 0LT, UK voice: +44 (0) 208 444 2081 mobile: +44 (0) 740 318 1612 |
Re: compound variables - kruskall wallis
|
Diana, you are not correct for Kruskal-Wallis. If you were right
about the necessity of equality-of-shapes assumption K-W
would have been useless 95% of time. Hopefully, in its general formulation,
K-W, like any nonparametric test, requires no distributional
assumptions. E.g. http://stats.stackexchange.com/q/76059/3277
K-W is by the way quite close in results (p-value) to ordinal logistic regression with default settings (i.e. logit link, no scale parameter). 20.11.2013 13:02, Kornbrot, Diana
пишет:
|
|
|
In reply to this post by Kornbrot, Diana
Hi Diana Thanks for your help. Sorry not much time to reply at the mo.
yes the compound variable is the combination of about six scale questions, combined in SPSS through 'data/compute variable. None of the variables are normally distributed so that's why I was investigating thing such as KW. Thanks for all contributions on this - I will read in detail later. Great to have this level of expertise to tap into. Mike On 20 November 2013 09:02, Kornbrot, Diana <[hidden email]> wrote:
|
Re: compound variables - kruskall wallis
|
|
As Bruce mentioned the
1950's brouhaha about nonparametrics still is causing
problems.
If you have 6 items that are intended to form a scale normality of items is rarely an issue. Severe discrepancy from normality of their mean or sum is vastly rarer. How did you obtain the the items for the scale? Did you write them, are they a well-established scale, a previously used scale but not photometrically checked out, etc.? Did you check the internal consistency reliability of your summative scale? How did you get the response scale for the frequency variable? Art Kendall Social Research ConsultantsOn 11/20/2013 7:39 AM, Oscar M [via SPSSX Discussion] wrote:
Art Kendall
Social Research Consultants |
Re: compound variables - kruskall wallis
|
In reply to this post by Oscar M
Mike Start further back. I prefer to use syntax rather than the GUI so, assuming your items are all coded in the same direction, something like (untested): File > new > syntax: Freq v1 to v6 / for not /bar. Corr v1 to v6. Then go for the fancy stuff: Reliability /scale (x) = var v1 to v6. [see page 1597 of the manual: Help > command syntax reference: nothing there about K-W] John F Hall (Mr) [Retired academic survey researcher] Email: [hidden email] Website: www.surveyresearch.weebly.com SPSS start page: www.surveyresearch.weebly.com/spss-without-tears.html From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of mike O'Driscoll Hi Diana Thanks for your help. Sorry not much time to reply at the mo. Mike On 20 November 2013 09:02, Kornbrot, Diana <[hidden email]> wrote: HI Thanks Diana
Professor Diana Kornbrot |
| Free forum by Nabble | Edit this page |