validating a questionnaire
Locked
Re: Comparing standard deviations
 
Administrator
|
That is what I recall about the CV too.
Here's what you said in your original post: "I have several tests which have different number of questions (from 20 to 100) so the maximum possible score on each test form is different." I gather that test scores are sums, is that right? Are the items on the tests all of the same form -- e.g., are they all 5- or 7-point Likert-type items? If so, why not just use means rather than sums as the test scores?
--
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/). |
% of varinace explained by factors
|
In reply to this post by DataMaestro
|
|
|
Dear All,
May be a silly question. My factor analysis came up with 16 factors/components out of 36 variables.I am going to do wealth index scoring through percentile ranking. Which factor out of 16 should I select and how? Thanking in advance, Khaing Soe M&E Officer UNICEF Myanmar ===================== 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 |
|
|
Khaing Soe,
The question is not silly; it's a nice practical question. I don't know anything about wealth indexes but I'm sure you do and so you will have read about the variables that make up the wealth indexes used in other countries. That is some guidance. Wealth score, which I'm assuming is a person level score rather than a country/region/district score, correlates with other things. What I'm not exactly sure but, again, the literture will offer guidance. If you measured some of those other things, you can compute your own set of correlations. Although you didn't ask for this, I'd like to comment on your factor analysis results. It seems to me that 16 factors from 36 items is way too many. On average, you have two items per factor. I'd guess that you have a number of single item factors. Think about the idea of a factor. An unobserved variable that strongly predicts a set of observed items (variables) that share a conceptual, logical, and semantic commonality. Gene Maguin >>May be a silly question. My factor analysis came up with 16 factors/components out of 36 variables.I am going to do wealth index scoring through percentile ranking. Which factor out of 16 should I select and how? Thanking in advance, Khaing Soe M&E Officer UNICEF Myanmar ===================== 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: % of variance explained by factors
|
|
In reply to this post by Humphrey Paulie
Since nobody else has responded to this, I will indulge in some observations. [I will use ‘cases’ and ‘responses’ interchangeably; and ‘items’ and ‘questions’ interchangeably.] Firstly, I trust that your data are good enough for factor analysis. For example,
If your intention was to create tests where all the items measure the same underlying concept, then I suppose a one-factor solution would seem to be comforting. I suppose the amount of variance explained by the first factor could be taken as a measure of how much the items do all assess the same thing. But presumably you would expect to see this figure reduce the more items you have. And the reason for the bullets above is that failure to find more than one factor might be an artefact of failing to meet the assumptions, especially on sample size. Low loading on that factor might also be due to failure of assumptions. However, if what you want to measure is the internal consistency of the test, surely Cronbach’s alpha would be the conventional way of doing it. (This suffers from the opposite problem in relation to items, i.e. Cronbach’s alpha tends to increase as the number of items increases.) If you are in a position to regard your questionnaires as exploratory, one way in which you could use your factor analysis results is to look for outlier items, i.e. questions which have low Squared Multiple Correlations in the initial solution [i.e. initial communalities in factor analysis (not in PCA, where they will all be 1)]. The variance in any such question does not overlap much with the other questions, so it does not appear to be measuring the same thing as the others. All of the foregoing only applies to reliability, i.e. how much the test is consistently measuring the same thing. There isn’t anything internal to the test which can measure validity, i.e. whether what you are measuring is what you intended to measure. This is my first post on this forum so I stand ready to be shot down! Date: Mon, 7 Jun 2010 12:54:22 -0700 From: [hidden email] Subject: % of varinace explained by factors To: [hidden email]
Get a free e-mail account with Hotmail. Sign-up now. |
| Free forum by Nabble | Edit this page |