question about associations between variables missing response categories
Locked
question about associations between variables missing response categories
 
|
Hi listserv,
I have a variable on a 0-10 scale (q2), and I would like to look at associations between this variable and 3 other variables (Q4-Q6). Q4-Q6 are all (in theory) on a scale of 1-4. However, in the actual data, the variable q2 only takes 4 values: 0, 7, 8 and 9. Q4-Q6 also has missing categories, for example Q4 only has values of 3 & 4. I would like to look at the association between q2 and each of the other 3 variables. Does anyone have any advice on this best way to do this? Specific transformations, specific statistical tests? Many thanks in advance for any assistance, Caroline ===================== 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: question about associations between variables missing response categories
|
|
Rather than 'missing', wouldn't 'unused' be a more accurate adjective. You want association. Why not correlations? They're computable. But since correlation assumes linearity why not begin with a crosstabs to whether the percentage of 4's rises or falls across values of Q2. You could also use Means (Q4 to Q6 by Q2) and test for linearity.
Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Caroline Wilson Sent: Wednesday, April 02, 2014 8:34 PM To: [hidden email] Subject: question about associations between variables missing response categories Hi listserv, I have a variable on a 0-10 scale (q2), and I would like to look at associations between this variable and 3 other variables (Q4-Q6). Q4-Q6 are all (in theory) on a scale of 1-4. However, in the actual data, the variable q2 only takes 4 values: 0, 7, 8 and 9. Q4-Q6 also has missing categories, for example Q4 only has values of 3 & 4. I would like to look at the association between q2 and each of the other 3 variables. Does anyone have any advice on this best way to do this? Specific transformations, specific statistical tests? Many thanks in advance for any assistance, Caroline ===================== 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 |
Re: question about associations between variables missing response categories
|
|
Thank you, Eugene. How about using Spearman’s Rho rank-order for correlation for ordinal variables, followed by ordinal regression (per Diana's suggestion)? I assume that q2 should be treated as ordinal - even though it is a 0-10 scale, only 4 of the values were actually used in the data. (?)
Thank you! Caroline -------------------------------------------- On Thu, 4/3/14, Maguin, Eugene <[hidden email]> wrote: Subject: Re: question about associations between variables missing response categories To: [hidden email] Date: Thursday, April 3, 2014, 9:04 AM Rather than 'missing', wouldn't 'unused' be a more accurate adjective. You want association. Why not correlations? They're computable. But since correlation assumes linearity why not begin with a crosstabs to whether the percentage of 4's rises or falls across values of Q2. You could also use Means (Q4 to Q6 by Q2) and test for linearity. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Caroline Wilson Sent: Wednesday, April 02, 2014 8:34 PM To: [hidden email] Subject: question about associations between variables missing response categories Hi listserv, I have a variable on a 0-10 scale (q2), and I would like to look at associations between this variable and 3 other variables (Q4-Q6).� Q4-Q6 are all (in theory) on a scale of 1-4. However, in the actual data, the variable q2 only takes 4 values: 0, 7, 8 and 9. Q4-Q6 also has missing categories, for example Q4 only has values of 3 & 4. I would like to look at the association between q2 and each of the other 3 variables. Does anyone have any advice on this best way to do this? Specific transformations, specific statistical tests? Many thanks in advance for any assistance, Caroline ===================== 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 ===================== 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: question about associations between variables missing response categories
|
|
I didn’t see Diana's reply on the list. Other than knowing that Spearman's is a rank order method, I'm also not very familiar with it. You're after association. I think the most general method is a chi-square or other method that treats the variables as nominal. I can't comment at all on those methods. It seems you'd like to assert that there's an ordinal or, maybe, linear relationship between q2 and q4, q5, and q6. So crosstab the variables and ask for Spearmans. The thing, the important thing, to see from the crosstabs is whether the proportion of 4's in q2 rises steadily (or falls steadily) across increasing values of the other variables. Let's take q2 and q4, specifically. You could treat q2 as the dv in a logistic regression but you get two coefficients for three values of q4. Or, you could treat q4 as the dv in an ordinal regression. You get one coefficient; however, the assumption is that slope coefficient is the same, i.e, not statistically different, for the two embedded regressions, which is testable if you use Plum but not, I don't think, if you use genlin or genlinmixed. You can get an idea of whether this might be from the crosstabs by computing the odds ratios.
Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Caroline Wilson Sent: Thursday, April 03, 2014 2:13 PM To: [hidden email] Subject: Re: question about associations between variables missing response categories Thank you, Eugene. How about using Spearman’s Rho rank-order for correlation for ordinal variables, followed by ordinal regression (per Diana's suggestion)? I assume that q2 should be treated as ordinal - even though it is a 0-10 scale, only 4 of the values were actually used in the data. (?) Thank you! Caroline -------------------------------------------- On Thu, 4/3/14, Maguin, Eugene <[hidden email]> wrote: Subject: Re: question about associations between variables missing response categories To: [hidden email] Date: Thursday, April 3, 2014, 9:04 AM Rather than 'missing', wouldn't 'unused' be a more accurate adjective. You want association. Why not correlations? They're computable. But since correlation assumes linearity why not begin with a crosstabs to whether the percentage of 4's rises or falls across values of Q2. You could also use Means (Q4 to Q6 by Q2) and test for linearity. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Caroline Wilson Sent: Wednesday, April 02, 2014 8:34 PM To: [hidden email] Subject: question about associations between variables missing response categories Hi listserv, I have a variable on a 0-10 scale (q2), and I would like to look at associations between this variable and 3 other variables (Q4-Q6). Q4-Q6 are all (in theory) on a scale of 1-4. However, in the actual data, the variable q2 only takes 4 values: 0, 7, 8 and 9. Q4-Q6 also has missing categories, for example Q4 only has values of 3 & 4. I would like to look at the association between q2 and each of the other 3 variables. Does anyone have any advice on this best way to do this? Specific transformations, specific statistical tests? Many thanks in advance for any assistance, Caroline ===================== 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 ===================== 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 |
Re: question about associations between variables missing response categories
|
|
In reply to this post by Caroline Wilson
[posting: third try, this time through Nabble]
Here is a general rule: simple rank transformation/statistics on rating scales with only a few points should be avoided, almost always. If there is no gap in the scores, you almost always do better by using the original numbers. Using ranks as an intermediate step in computing some logistic solution , on the other hand, can be sensible and powerful. This example does have a 10-point scale with a huge (unexplained) gap. (Why?) Instead of the Spearman: The Crosstabs procedure provides the Mantel statistic, which cleverly treats the categories as if they were coded without a gap. WHY is there a multiple-point gap in q2? Should q2 be treated, for purposes of inference, as two questions -- (0 vs. other) and (points, for the non-zero)? Alternatively: Should the 0 be recoded to be contiguous? Whatever you do, keep in mind that correlations generated by 2-point items are not "robust" when compared to those from 3-point or 4-point items. That is to say: a Spearman or Pearson of 0.40 (say) on a dichotomy might reflect an underlying relation that is stronger than the parallel measure of 0.45 for a 4-point item, -- Rich Ulrich |
Re: question about associations between variables missing response categories
|
|
Thank you, Rich and Gene. Your advice is invaluable!
-------------------------------------------- On Thu, 4/3/14, Rich Ulrich <[hidden email]> wrote: Subject: Re: question about associations between variables missing response categories To: [hidden email] Date: Thursday, April 3, 2014, 4:31 PM [posting: third try, this time through Nabble] Here is a general rule: simple rank transformation/statistics on rating scales with only a few points should be avoided, almost always. If there is no gap in the scores, you almost always do better by using the original numbers. Using ranks as an intermediate step in computing some logistic solution , on the other hand, can be sensible and powerful. This example does have a 10-point scale with a huge (unexplained) gap. (Why?) Instead of the Spearman: The Crosstabs procedure provides the Mantel statistic, which cleverly treats the categories as if they were coded without a gap. WHY is there a multiple-point gap in q2? Should q2 be treated, for purposes of inference, as two questions -- (0 vs. other) and (points, for the non-zero)? Alternatively: Should the 0 be recoded to be contiguous? Whatever you do, keep in mind that correlations generated by 2-point items are not "robust" when compared to those from 3-point or 4-point items. That is to say: a Spearman or Pearson of 0.40 (say) on a dichotomy might reflect an underlying relation that is stronger than the parallel measure of 0.45 for a 4-point item, -- Rich Ulrich -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/question-about-associations-between-variables-missing-response-categories-tp5725206p5725253.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 |
Re: question about associations between variables missing response categories
|
|
In reply to this post by Rich Ulrich
Rich, in response to your question about why there is such a large gap in q2: I'm not sure there was a reason other than simply that respondents just did not select those other values when completing the survey.
-------------------------------------------- On Thu, 4/3/14, Rich Ulrich <[hidden email]> wrote: Subject: Re: question about associations between variables missing response categories To: [hidden email] Date: Thursday, April 3, 2014, 4:31 PM [posting: third try, this time through Nabble] Here is a general rule: simple rank transformation/statistics on rating scales with only a few points should be avoided, almost always. If there is no gap in the scores, you almost always do better by using the original numbers. Using ranks as an intermediate step in computing some logistic solution , on the other hand, can be sensible and powerful. This example does have a 10-point scale with a huge (unexplained) gap. (Why?) Instead of the Spearman: The Crosstabs procedure provides the Mantel statistic, which cleverly treats the categories as if they were coded without a gap. WHY is there a multiple-point gap in q2? Should q2 be treated, for purposes of inference, as two questions -- (0 vs. other) and (points, for the non-zero)? Alternatively: Should the 0 be recoded to be contiguous? Whatever you do, keep in mind that correlations generated by 2-point items are not "robust" when compared to those from 3-point or 4-point items. That is to say: a Spearman or Pearson of 0.40 (say) on a dichotomy might reflect an underlying relation that is stronger than the parallel measure of 0.45 for a 4-point item, -- Rich Ulrich -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/question-about-associations-between-variables-missing-response-categories-tp5725206p5725253.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 |
Re: question about associations between variables missing response categories
|
How many respondents do
you have?
What are the value labels for the responses? Art Kendall Social Research ConsultantsOn 4/3/2014 5:06 PM, Caroline Wilson [via SPSSX Discussion] wrote: Rich, in response to your question about why there is such a large gap in q2: I'm not sure there was a reason other than simply that respondents just did not select those other values when completing the survey.
Art Kendall
Social Research Consultants |
Re: question about associations between variables missing response categories
|
|
Hi Art - there are 75 respondents. For q2, the 0-10 relates to degree of effort. For the other questions, the value labels are definitely disagree, somewhat disagree, somewhat agree, definitely agree.
-------------------------------------------- On Thu, 4/3/14, Art Kendall <[hidden email]> wrote: Subject: Re: question about associations between variables missing response categories To: [hidden email] Date: Thursday, April 3, 2014, 5:35 PM How many respondents do you have? What are the value labels for the responses? Art Kendall Social Research Consultants On 4/3/2014 5:06 PM, Caroline Wilson [via SPSSX Discussion] wrote: Rich, in response to your question about why there is such a large gap in q2: I'm not sure there was a reason other than simply that respondents just did not select those other values when completing the survey. -------------------------------------------- On Thu, 4/3/14, Rich Ulrich <[hidden email]> wrote: Subject: Re: question about associations between variables missing response categories To: [hidden email] Date: Thursday, April 3, 2014, 4:31 PM [posting: third try, this time through Nabble] Here is a general rule: simple rank transformation/statistics on rating scales with only a few points should be avoided, almost always. If there is no gap in the scores, you almost always do better by using the original numbers. Using ranks as an intermediate step in computing some logistic solution , on the other hand, can be sensible and powerful. This example does have a 10-point scale with a huge (unexplained) gap. (Why?) Instead of the Spearman: The Crosstabs procedure provides the Mantel statistic, which cleverly treats the categories as if they were coded without a gap. WHY is there a multiple-point gap in q2? Should q2 be treated, for purposes of inference, as two questions -- (0 vs. other) and (points, for the non-zero)? Alternatively: Should the 0 be recoded to be contiguous? Whatever you do, keep in mind that correlations generated by 2-point items are not "robust" when compared to those from 3-point or 4-point items. That is to say: a Spearman or Pearson of 0.40 (say) on a dichotomy might reflect an underlying relation that is stronger than the parallel measure of 0.45 for a 4-point item, -- Rich Ulrich -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/question-about-associations-between-variables-missing-response-categories-tp5725206p5725253.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 If you reply to this email, your message will be added to the discussion below: http://spssx-discussion.1045642.n5.nabble.com/question-about-associations-between-variables-missing-response-categories-tp5725206p5725258.html To start a new topic under SPSSX Discussion, email [hidden email] To unsubscribe from SPSSX Discussion, click here. NAML Art Kendall Social Research Consultants View this message in context: Re: question about associations between variables missing response categories Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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: question about associations between variables missing response categories
|
|
given the other responses
and the context does it make sense that those respondents would
put not effort into something, e.g., that they do not do
whatever the effort is applied to?
Is it possible that the question stem was ambiguous? Art Kendall Social Research ConsultantsOn 4/3/2014 5:47 PM, Caroline Wilson [via SPSSX Discussion] wrote: Hi Art - there are 75 respondents. For q2, the 0-10 relates to degree of effort. For the other questions, the value labels are definitely disagree, somewhat disagree, somewhat agree, definitely agree.
Art Kendall
Social Research Consultants |
Re: question about associations between variables missing response categories
|
|
I think a value of 0 on the 0-10 scale could mean "no effort." there are only 2 respondents who selected 0.
It is possible that q2 was not an entirely clear question. thanks! -------------------------------------------- On Thu, 4/3/14, Art Kendall <[hidden email]> wrote: Subject: Re: question about associations between variables missing response categories To: [hidden email] Date: Thursday, April 3, 2014, 6:11 PM given the other responses and the context does it make sense that those respondents would put not effort into something, e.g., that they do not do whatever the effort is applied to? Is it possible that the question stem was ambiguous? Art Kendall Social Research Consultants On 4/3/2014 5:47 PM, Caroline Wilson [via SPSSX Discussion] wrote: Hi Art - there are 75 respondents. For q2, the 0-10 relates to degree of effort. For the other questions, the value labels are definitely disagree, somewhat disagree, somewhat agree, definitely agree. -------------------------------------------- On Thu, 4/3/14, Art Kendall <[hidden email]> wrote: Subject: Re: question about associations between variables missing response categories To: [hidden email] Date: Thursday, April 3, 2014, 5:35 PM How many respondents do you have? What are the value labels for the responses? Art Kendall Social Research Consultants On 4/3/2014 5:06 PM, Caroline Wilson [via SPSSX Discussion] wrote: Rich, in response to your question about why there is such a large gap in q2: I'm not sure there was a reason other than simply that respondents just did not select those other values when completing the survey. -------------------------------------------- On Thu, 4/3/14, Rich Ulrich <[hidden email]> wrote: Subject: Re: question about associations between variables missing response categories To: [hidden email] Date: Thursday, April 3, 2014, 4:31 PM [posting: third try, this time through Nabble] Here is a general rule: simple rank transformation/statistics on rating scales with only a few points should be avoided, almost always. If there is no gap in the scores, you almost always do better by using the original numbers. Using ranks as an intermediate step in computing some logistic solution , on the other hand, can be sensible and powerful. This example does have a 10-point scale with a huge (unexplained) gap. (Why?) Instead of the Spearman: The Crosstabs procedure provides the Mantel statistic, which cleverly treats the categories as if they were coded without a gap. WHY is there a multiple-point gap in q2? Should q2 be treated, for purposes of inference, as two questions -- (0 vs. other) and (points, for the non-zero)? Alternatively: Should the 0 be recoded to be contiguous? Whatever you do, keep in mind that correlations generated by 2-point items are not "robust" when compared to those from 3-point or 4-point items. That is to say: a Spearman or Pearson of 0.40 (say) on a dichotomy might reflect an underlying relation that is stronger than the parallel measure of 0.45 for a 4-point item, -- Rich Ulrich -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/question-about-associations-between-variables-missing-response-categories-tp5725206p5725253.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 If you reply to this email, your message will be added to the discussion below: http://spssx-discussion.1045642.n5.nabble.com/question-about-associations-between-variables-missing-response-categories-tp5725206p5725258.html To start a new topic under SPSSX Discussion, email [hidden email] To unsubscribe from SPSSX Discussion, click here. NAML Art Kendall Social Research Consultants View this message in context: Re: question about associations between variables missing response categories Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 If you reply to this email, your message will be added to the discussion below: http://spssx-discussion.1045642.n5.nabble.com/question-about-associations-between-variables-missing-response-categories-tp5725206p5725262.html To start a new topic under SPSSX Discussion, email [hidden email] To unsubscribe from SPSSX Discussion, click here. NAML Art Kendall Social Research Consultants View this message in context: Re: question about associations between variables missing response categories Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |