After the fact problem
Locked
 
|
Dear Listers,
Does this solution work? I had administered a survey that looks at the relationship between well-being and balance of life domains. Well-being, the outcome, was measure with Ryff's Psychological Well Being Scale. Life balance was defined as the standard deviation of time spent in ten 10 life domains. The survey asked how much time per week the participant spent doing activities in each of ten domains (e.g., Health and self-maintenance, Spirituality or religion, school, etc..). A sample question was: How many hours, per week, do you spend with, or in service to, a spiritual community (including attending religious services)?______hours per week. The problem is the survey's instructions did not limit the total hours allocated to 168. Thus, I got total hours ranging from 125 to 435. One solution was to divide each domain time by the total time for the participant. That is, for each participant, to convert domain times into percentage of their total time. I then plan to regress the well-being measure(s) on these ten converted domains. Am I introducing an artifact into the analysis with the dividing by the participants' total time? TIA, Stephen Salbod, Pace University, NYC PS Anyone have a time machine to lend :) ===================== 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 |
Administrator
|
I guess some people are excellent at multi-tasking and don't require sleep?
I have no solution but just couldn't resist making a comment ;-)
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me. --- "Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis." Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?" |
|
|
In reply to this post by Salbod
As a practical matter I don't see any other
good alternative. Are the really high or really low numbers of hours
outliers that could perhaps be dropped? Is there any possibility
that the people who gave less than 168 hours spend some time doing things
they didn't consider to fall into the 10 domains you specified?
Martha Hewett From: "Salbod, Mr. Stephen" <[hidden email]> To: [hidden email] Date: 08/24/2012 09:37 AM Subject: After the fact problem Sent by: "SPSSX(r) Discussion" <[hidden email]> Dear Listers, Does this solution work? I had administered a survey that looks at the relationship between well-being and balance of life domains. Well-being, the outcome, was measure with Ryff's Psychological Well Being Scale. Life balance was defined as the standard deviation of time spent in ten 10 life domains. The survey asked how much time per week the participant spent doing activities in each of ten domains (e.g., Health and self-maintenance, Spirituality or religion, school, etc..). A sample question was: How many hours, per week, do you spend with, or in service to, a spiritual community (including attending religious services)?______hours per week. The problem is the survey's instructions did not limit the total hours allocated to 168. Thus, I got total hours ranging from 125 to 435. One solution was to divide each domain time by the total time for the participant. That is, for each participant, to convert domain times into percentage of their total time. I then plan to regress the well-being measure(s) on these ten converted domains. Am I introducing an artifact into the analysis with the dividing by the participants' total time? TIA, Stephen Salbod, Pace University, NYC PS Anyone have a time machine to lend :) ===================== 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 |
|
|
Bear in mind that the normalized sum will
be constant, so if you are using REGRESSION, you need to suppress the constant
term, i.e., you really only have 9 independent measures.
Jon Peck (no "h") aka Kim Senior Software Engineer, IBM [hidden email] new phone: 720-342-5621 From: Martha Hewett <[hidden email]> To: [hidden email] Date: 08/24/2012 09:44 AM Subject: Re: [SPSSX-L] After the fact problem Sent by: "SPSSX(r) Discussion" <[hidden email]> As a practical matter I don't see any other good alternative. Are the really high or really low numbers of hours outliers that could perhaps be dropped? Is there any possibility that the people who gave less than 168 hours spend some time doing things they didn't consider to fall into the 10 domains you specified? Martha Hewett From: "Salbod, Mr. Stephen" <[hidden email]> To: [hidden email] Date: 08/24/2012 09:37 AM Subject: After the fact problem Sent by: "SPSSX(r) Discussion" <[hidden email]> Dear Listers, Does this solution work? I had administered a survey that looks at the relationship between well-being and balance of life domains. Well-being, the outcome, was measure with Ryff's Psychological Well Being Scale. Life balance was defined as the standard deviation of time spent in ten 10 life domains. The survey asked how much time per week the participant spent doing activities in each of ten domains (e.g., Health and self-maintenance, Spirituality or religion, school, etc..). A sample question was: How many hours, per week, do you spend with, or in service to, a spiritual community (including attending religious services)?______hours per week. The problem is the survey's instructions did not limit the total hours allocated to 168. Thus, I got total hours ranging from 125 to 435. One solution was to divide each domain time by the total time for the participant. That is, for each participant, to convert domain times into percentage of their total time. I then plan to regress the well-being measure(s) on these ten converted domains. Am I introducing an artifact into the analysis with the dividing by the participants' total time? TIA, Stephen Salbod, Pace University, NYC PS Anyone have a time machine to lend :) ===================== 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 |
|
|
In reply to this post by Salbod
"the problem is..." -- Why, exactly, is "not-168-hours" a problem?
The biggest problem seems to be that you (or someone) wrongly expected that the numbers total to 168; and, thus, you demonstrate that the scale was created in a sloppy enough fashion that not everyone uses the same basis for answering it. (Unless you are the only one who is out of step.) It does seem pretty natural to put some activities into more than one category -- Playing ball is both exercise and social. Filling out the numbers as exclusive categories requires more rules, or makes people make their own guesses of what is intended. Having different numbers of hours per person avoids certain difficulties that arise when you *do* have predictors that add to a total. What I'm really curious about is that the outcome is the SD of the times, regarded as Well-being. I would think that a label like that would have Good at one end and Poor at the other. The smallest SD would probably go to the person who has least Sleep (assuming that the total of 168 was meaningful). Is this supposed to be Good or Poor for Well-being? -- Rich Ulrich > Date: Fri, 24 Aug 2012 14:29:39 +0000 > From: [hidden email] > Subject: After the fact problem > To: [hidden email] > > Dear Listers, > > Does this solution work? > > I had administered a survey that looks at the relationship between well-being and balance of life domains. Well-being, the outcome, was measure with Ryff's Psychological Well Being Scale. Life balance was defined as the standard deviation of time spent in ten 10 life domains. The survey asked how much time per week the participant spent doing activities in each of ten domains (e.g., Health and self-maintenance, Spirituality or religion, school, etc..). A sample question was: > > How many hours, per week, do you spend with, or in service to, a spiritual community (including attending religious services)?______hours per week. > > The problem is the survey's instructions did not limit the total hours allocated to 168. Thus, I got total hours ranging from 125 to 435. > One solution was to divide each domain time by the total time for the participant. That is, for each participant, to convert domain times into percentage of their total time. > > I then plan to regress the well-being measure(s) on these ten converted domains. > > Am I introducing an artifact into the analysis with the dividing by the participants' total time? > > ... |
|
|
Rich, everyone seems to have their own idea about the number of hours in a week. For example,
the first 5 participants in Domain-1 could conceivably be: 5, 5, 5, 5, and 5. But, their total week hours could 10, 60, 40, 80, and 200. I could not say that
the first 5 participants spent the same amount of time in Domain-1. I hope that makes sense.
From: Rich Ulrich [mailto:[hidden email]]
"the problem is..." -- Why, exactly, is "not-168-hours" a problem? > Date: Fri, 24 Aug 2012 14:29:39 +0000 |
|
|
In reply to this post by Jon K Peck
Thank you, Jon, for your insightful response. I’m not clear on how to go about my regression analysis after reading your recommendation.
Are you suggesting that I drop a Domain(e.g., perDom10) and include /ORIGIN? i.e.: REGRESSION /MISSING LISTWISE /STATISTICS COEFF OUTS R ANOVA ZPP /CRITERIA=PIN(.05) POUT(.10) /ORIGIN
/DEPENDENT pwbs /METHOD=ENTER perDom1 perDom2 perDom3 perDom4 perDom5 perDom6 perDom7 perDom8 perDom9. From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Jon K Peck Bear in mind that the normalized sum will be constant, so if you are using REGRESSION, you need to suppress the constant term, i.e., you really only have 9 independent measures.
|
|
|
In reply to this post by Salbod
Well, you have the advantage of having some data on hand.
But to my sensibility, each of the 5 who have 5 hours in domain-1 do have the same amount of time, 5 hours. That is the simplest and most direct reading of the data. You *can*, if you want, make some more abstract construction, including the prospect of using "compositional data." (Aitchison is a name that I saw before.) There are books written on analysis of "compositional data", which is what you have when it adds up to a fixed total. I looked into that a little bit, once -- I don't remember any discussion of having non-matched totals. Your shortfall of hours looks like some version of an incomplete-data or missing-data problem. You regularly need to make strong assumptions to deal with those. The excess-hours problem is what makes me think that using simple hours could be the fairest approach. I'm still curious about how one justifies a theoretical construct called "well-being" based on the standard deviation of hours in categories. -- Rich Ulrich Date: Fri, 24 Aug 2012 18:44:37 +0000 From: [hidden email] Subject: Re: After the fact problem To: [hidden email] Rich, everyone seems to have their own idea about the number of hours in a week. For example, the first 5 participants in Domain-1 could conceivably be: 5, 5, 5, 5, and 5. But, their total week hours could 10, 60, 40, 80, and 200. I could not say that the first 5 participants spent the same amount of time in Domain-1.
I hope that makes sense.
From: Rich Ulrich [mailto:[hidden email]]
"the problem is..." -- Why, exactly, is "not-168-hours" a problem? > Date: Fri, 24 Aug 2012 14:29:39 +0000 |
| Free forum by Nabble | Edit this page |