Importance of individual scores in contributing to a sum
Locked
Importance of individual scores in contributing to a sum
 
|
Dear Listers, I've been out of touch for some time, so please forgive me if something like this has come up before. I have a client who wants to conduct a survey in which respondents will report on various sources of information for learning about something. For example, it might be: Books Magazines Internet Friends TV The client has not yet decided whether each item will be rated either as Yes/No (scored as 1 or 0) or on a short numerical scale (Like Often=3, Sometimes=2 , Rarely=1, Never=0, or something a bit more sophisticated). Either way, the scores across each item will be summed for an overall score of how much information is sought. There is interest in the question as to the degree to which each item contributes most to the overall score. The literal answer to this is obvious: the item that is answered most often (if Yes/No is used) or has the highest mean score across respondents is the most important, the 2nd highest mean is 2nd most important, etc. But is there some sort of test of the statistical significance between these means? I am thinking of something like the difference between means in a 1-way ANOVA -- but in that case, the means are based on separate groups of people, whereas in this case it is multiple variables on the same set of people. It seems like this question should be fairly common, but I can't think of how it should be handled. Thanks for any ideas! Allan Research Consulting [hidden email] Business & Cell (any time): 215-820-8100 NEW Address: 587 Shotgun Spring Rd, New Market, VA 22844 Visit my Web site at www.dissertationconsulting.net ===================== 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 |
|
Whether each response is comprised of 2,3,4,5,…multiple choices, it appears that your best analysis options are: · Factor Analysis · Discriminant Analysis · Cluster Analysis · Several other neat analysis For example, using Principal Component analysis you can transfer a set of interrelated variable into a new set of uncorrelated components which account for all the variance in the original variables. Max. From: Allan Lundy, PhD [via SPSSX Discussion] [mailto:[hidden email]]
Allan Lundy, PhD If you reply to this email, your message will be added to the discussion below: To start a new topic under SPSSX Discussion, email [hidden email] |
Re: Importance of individual scores in contributing to a sum
|
|
In reply to this post by Allan Lundy, PhD
The test for two paired measures is the paired t-test. The test for several matched measures is the repeated measures ANOVA, looking at the test between measures. If this test does show an overall difference, the follow-up contrasts between pairs are often wisely performed as (again) paired t-tests, to allow for differences in variances and correlations. This is a place where the Benjamini correction for multiple tests might be used, False Discovery Rate ... but be careful to use the original version rather than one of the modifications proposed for other sorts of circumstances. As another poster suggests - If you can reduce your large number of measures to make logical composite scores, a-priori or after factor analysis, you probably will leave yourself in a better position for drawing conclusions about what matters. -- Rich Ulrich Date: Wed, 5 Mar 2014 20:14:12 -0500 From: [hidden email] Subject: Importance of individual scores in contributing to a sum To: [hidden email] Dear Listers, I've been out of touch for some time, so please forgive me if something like this has come up before. I have a client who wants to conduct a survey in which respondents will report on various sources of information for learning about something. For example, it might be: Books Magazines Internet Friends TV The client has not yet decided whether each item will be rated either as Yes/No (scored as 1 or 0) or on a short numerical scale (Like Often=3, Sometimes=2 , Rarely=1, Never=0, or something a bit more sophisticated). Either way, the scores across each item will be summed for an overall score of how much information is sought. There is interest in the question as to the degree to which each item contributes most to the overall score. The literal answer to this is obvious: the item that is answered most often (if Yes/No is used) or has the highest mean score across respondents is the most important, the 2nd highest mean is 2nd most important, etc. But is there some sort of test of the statistical significance between these means? I am thinking of something like the difference between means in a 1-way ANOVA -- but in that case, the means are based on separate groups of people, whereas in this case it is multiple variables on the same set of people. It seems like this question should be fairly common, but I can't think of how it should be handled. Thanks for any ideas! Allan Allan Lundy, PhD Research Consulting [hidden email] Business & Cell (any time): 215-820-8100<span class="skype_c2c_container" dir="ltr" tabindex="-1" onmouseover="SkypeClick2Call.MenuInjectionHandler.showMenu(this, event)" onmouseout="SkypeClick2Call.MenuInjectionHandler.hideMenu(event)" skype_menu_props="{"numberToCall":"+12158208100","isFreecall":false,"isMobile":false,"isRtl":false}"><img class="skype_c2c_logo_img" src="resource://skype_ff_extension-at-jetpack/skype_ff_extension/data/call_skype_logo.png">215-820-8100 NEW Address: 587 Shotgun Spring Rd, New Market, VA 22844 Visit my Web site at www.dissertationconsulting.net ===================== 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 You'll need Skype CreditFree via Skype |
Re: Importance of individual scores in contributing to a sum
|
In reply to this post by Allan Lundy, PhD
Rich has given you a
useful answer about what to do with the numbers you obtain.
I suggest that you you use as many response values as your respondents deal with. Since the underlying construct is continuous more values will more closely represent that construct. The more restricted the range of valid values in a variable the lower the maximum possible correlation wit other variables. The corrected item-total correlation from RELIABILITY will tell you how well each item correlated with the sum of the other items. The squared multiple correlation from RELIABILITY will give you a similar perspective. Be sure to look at the internal consistency when the items are standardized. Art Kendall Social Research ConsultantsOn 3/5/2014 8:27 PM, Allan Lundy, PhD [via SPSSX Discussion] wrote:
Art Kendall
Social Research Consultants |
|
|
In reply to this post by Allan Lundy, PhD
Summing the items seems fruitless to me, given the objective and description of the items. Perhaps pairwise comparisons will answer your questions...Modeling multiple ordered ratings items might lead you to a multivariate ordinal regression.
Want a more sophisticated analysis? Consider a latent class analysis to uncover topologies of individuals with respect to their pattern(s) to obtain information. Inclusion of covariates could enrich such an analysis. Ryan ===================== 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 |
| Free forum by Nabble | Edit this page |