Comparing ordinal variable in k Independant samples
Locked
Comparing ordinal variable in k Independant samples
 
|
Hello to everybody: I need to check for differences among four independent samples that made a short essay (rated using a 3 point rating scale= good, medium, poor). I presume I could use Kruskal-Wallis H or a similar non-parametric test, but in the new PASW 18 non-parametric test “Test Field” option only accepts variable measured in scale level. Can’t I use ordinal variables? What test could I do?
Thanks
|
Re: Comparing ordinal variable in k Independant samples
Administrator
|
Ordinal logistic regression (via the PLUM procedure) might be suitable. How large is your sample? And how many subjects fall into each of the 3 categories? If your data don't meet all the assumptions for ordinal logistic regression, you could try multinomial logistic regression (NOMREG) instead. It assumes only nominal scale measurement for the outcome, not ordered categories.
--
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/). |
|
|
Hello,
I am having a low IQ day today. I need to compare the proportion of males vs females on a series of survey question. The substantive questions is is there a gender effect (i.e., are the proportions significantly different). The rub is I do not have the raw data, only cross tabulation tables.
My first thought is a simple t-test of proportions or would a chi-square test of fit be more appropriate (where the observed is %male and the expected is %female)??
Thanks in advance
John |
Re: easy non-SPSS question (analysis)
|
Administrator
|
Use the WEIGHT command. Here's an example. Variable COUNT holds the cell counts. Adjust the number of categories for the other variable as needed.
DATA LIST LIST /sex (f2.0) cat (f2.0) count(f5.0) . BEGIN DATA. 1 1 9 1 2 17 1 3 22 2 1 21 2 2 13 2 3 8 END DATA. val lab sex 1 'Male' 2 'Female' / cat 1 'Category 1' 2 'Category 2' 3 'Category 3' . weight by count. crosstabs sex by cat / stat = chisq.
--
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: easy non-SPSS question (analysis)
|
|
In reply to this post by J P-6
The PROPOR extension command from SPSS Developer Central (www.spss.com/devcentral) displays binomial and Poisson c.i.'s for proportions. It can be used when you only have the counts with the data either specified in the command or in the usual way as variables. Usage example: PROPOR NUM=55 DENOM=100. NUM and DENOM can also be lists of values. PROPOR /HELP. displays the full syntax information. HTH, Jon Peck SPSS, an IBM Company [hidden email] 312-651-3435
Hello, I am having a low IQ day today. I need to compare the proportion of males vs females on a series of survey question. The substantive questions is is there a gender effect (i.e., are the proportions significantly different). The rub is I do not have the raw data, only cross tabulation tables. My first thought is a simple t-test of proportions or would a chi-square test of fit be more appropriate (where the observed is %male and the expected is %female)?? Thanks in advance John |
Re: easy non-SPSS question (analysis)
|
Administrator
|
These confidence intervals may be interesting and useful, but are not equivalent to a test on whether the proportions for males & females differ. I.e., overlap in the confidence intervals can occur even if the difference between the point estimates is statistically significant. E.g.,
http://www.cmaj.ca/cgi/content/full/166/1/65 Cheers, Bruce
--
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: Comparing ordinal variable in k Independant samples
|
|
In reply to this post by Bruce Weaver
or, if you have the CATEGORIES module, you could use CATREG (Analyze menu, Regression, Optimal Scaling).
Kind regards,
Anita van der Kooij,
Data Theory Group,
Leiden University. From: SPSSX(r) Discussion on behalf of Bruce Weaver Sent: Tue 15-Jun-10 19:54 To: [hidden email] Subject: Re: Comparing ordinal variable in k Independant samples Ordinal logistic regression (via the PLUM procedure) might be suitable. How ********************************************************************** This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify the system manager. **********************************************************************
|
SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples)
|
|
In reply to this post by ANDRES ALBERTO BURGA LEON
There was a question that I just stumbled across a month late, but I
wanted to bring it up again, not because the earlier answers were bad, but rather because it illustrates a philosophical change in how SPSS handles data analysis. I don't like this change and I wanted to see what others think about it. ANDRES ALBERTO BURGA LEON wrote: > I need to check for differences among four independent samples that made > a short essay (rated using a 3 point rating scale= good, medium, poor). > I presume I could use Kruskal-Wallis H or a similar non-parametric test, > but in the new PASW 18 non-parametric test “Test Field” option only > accepts variable measured in scale level. Can’t I use ordinal variables? > What test could I do? I just got SPSS 18 loaded last week, and I looked at the non-parametric test procedure. They've integrated all the nonparametric tests into a single menu choice, which may or may not be a good thing, but interestingly, this appears to be the first (but probably not the last) statistical procedure that uses the nominal/ordinal/scale properties of the variable. The graphical methods, of course, have been using this nominal/ordinal/scale property for quite a while (since version 15, I believe). This is something like the philosophy implemented in JMP, but it is, so far, only implemented partially in SPSS. I'm not sure I like this new approach and I thought it would be worth discussing this on this list. In theory, the measurement property of a particular variable should allow you to use or eliminate certain tests or graphs, but there are two problems with this. First, a lot of times, you want to run a test that doesn't quite meet the measurement properties of the variable in question just as a sensitivity check. Second, SPSS does a lousy job of assigning measurement properties to a file that is imported from another source. I dislike the idea of having to check and fix the measurement properties of every variable in every imported data set. The advantage is that incorporating measurement properties into all the statistical procedures might prevent an inexperienced data analyst from making a bogus choice and might end up steering them towards a more appropriate choice. It also is a potential time saver in that you will be presented with a smaller number of valid choices for your graphs and analyses. The workaround is to change the measurement properties temporarily, but I find this tedious and annoying. Another workaround is to use the legacy dialogs, which did not have these restrictions. I'm not thrilled about this either. I don't want to teach people to use something that is obsolete. I also see this as a common question that I will get in consulting (why doesn't SPSS let me run the X procedure on my data set). It might end up padding my consulting income, but I still don't like it. What do other people think about this use of measurement properties as a gatekeeper that prevents certain graphs/analyses from being run? Steve Simon, Standard Disclaimer "Data entry and data management issues with examples in IBM SPSS," Tuesday, August 24, 11am-noon CDT. Free webinar. Details at www.pmean.com/webinars ===================== 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 |
|
|
Dear List,
Is there a rule of thumb about how much skewness is acceptable vs. unacceptable. I have read conflicting opinions that say that skewness should be less than 2, and then some people have said that skewness should be less than 1. All suggestions are welcomed, Stace ===================== 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: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples)
|
|
In reply to this post by Steve Simon, P.Mean Consulting
I strongly agree with you Steve; and the argument for this system, i.e.
that it will stop users running tests using variables of the wrong type is weak in itself in that if users don't understand why their test is appropriate they are unlikely to be able to interpret the results properly anyway... I can also think of instances where using a method not normally associated with certain types of vars can be useful e.g. mean tables on a dichotomous variable will give you a proportion scoring 1. And if these restrictions contuinue to be implemented, will we be stopped from adding dummy variables to regression!? This is a recent change that definitely should be reversed (along with the loss of the ability to right-click on menu headings to get a brief help box) cheers Chris On 19/07/2010 18:51, Steve Simon, P.Mean Consulting wrote: > There was a question that I just stumbled across a month late, but I > wanted to bring it up again, not because the earlier answers were bad, > but rather because it illustrates a philosophical change in how SPSS > handles data analysis. I don't like this change and I wanted to see what > others think about it. > > ANDRES ALBERTO BURGA LEON wrote: > >> I need to check for differences among four independent samples that made >> a short essay (rated using a 3 point rating scale= good, medium, poor). >> I presume I could use Kruskal-Wallis H or a similar non-parametric test, >> but in the new PASW 18 non-parametric test “Test Field” option only >> accepts variable measured in scale level. Can’t I use ordinal variables? >> What test could I do? > > I just got SPSS 18 loaded last week, and I looked at the non-parametric > test procedure. They've integrated all the nonparametric tests into a > single menu choice, which may or may not be a good thing, but > interestingly, this appears to be the first (but probably not the last) > statistical procedure that uses the nominal/ordinal/scale properties of > the variable. The graphical methods, of course, have been using this > nominal/ordinal/scale property for quite a while (since version 15, I > believe). > > This is something like the philosophy implemented in JMP, but it is, so > far, only implemented partially in SPSS. I'm not sure I like this new > approach and I thought it would be worth discussing this on this list. > > In theory, the measurement property of a particular variable should > allow you to use or eliminate certain tests or graphs, but there are two > problems with this. First, a lot of times, you want to run a test that > doesn't quite meet the measurement properties of the variable in > question just as a sensitivity check. Second, SPSS does a lousy job of > assigning measurement properties to a file that is imported from another > source. I dislike the idea of having to check and fix the measurement > properties of every variable in every imported data set. > > The advantage is that incorporating measurement properties into all the > statistical procedures might prevent an inexperienced data analyst from > making a bogus choice and might end up steering them towards a more > appropriate choice. It also is a potential time saver in that you will > be presented with a smaller number of valid choices for your graphs and > analyses. > > The workaround is to change the measurement properties temporarily, but > I find this tedious and annoying. > > Another workaround is to use the legacy dialogs, which did not have > these restrictions. I'm not thrilled about this either. I don't want to > teach people to use something that is obsolete. > > I also see this as a common question that I will get in consulting (why > doesn't SPSS let me run the X procedure on my data set). It might end up > padding my consulting income, but I still don't like it. > > What do other people think about this use of measurement properties as a > gatekeeper that prevents certain graphs/analyses from being run? > > Steve Simon, Standard Disclaimer > "Data entry and data management issues with examples > in IBM SPSS," Tuesday, August 24, 11am-noon CDT. > Free webinar. Details at www.pmean.com/webinars > > ===================== > 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 |
|
Administrator
|
In reply to this post by stace swayne
Please provide some context. Too much skewness for what? Which measure of skewness are you talking about?
--
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: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples)
|
|
In reply to this post by Christopher Stride
I also agree strongly. There are times that variables that might be nominal in one situation may actually be interval/ratio in another context (e.g., Lord's treatise "On the Statistical Treatment of Football Numbers" from the early 1950s). The Steven's view of measurement is not universally accepted, with a number of psychometricians and statisticians arguing against the NOIR categorization. (I can post some of these references if anyone is interested.)
At bottom, my sense is that the software should not limit the researcher in any sort of mechanically enforced fashion. Those of us who know what we are doing should not have to go through the extra hoops that such a system imposes. My two cents' worth . . . Harley Dr. Harley Baker Professor and Chair, Psychology Program Sage Hall 2061 California State University Channel Islands One University Drive Camarillo, CA 93012 805.437.8997 (p) 805.437.8951 (f) [hidden email] ________________________________________ From: SPSSX(r) Discussion [[hidden email]] on behalf of Dr C B Stride [[hidden email]] Sent: Monday, July 19, 2010 12:04 PM To: [hidden email] Subject: Re: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples) I strongly agree with you Steve; and the argument for this system, i.e. that it will stop users running tests using variables of the wrong type is weak in itself in that if users don't understand why their test is appropriate they are unlikely to be able to interpret the results properly anyway... I can also think of instances where using a method not normally associated with certain types of vars can be useful e.g. mean tables on a dichotomous variable will give you a proportion scoring 1. And if these restrictions contuinue to be implemented, will we be stopped from adding dummy variables to regression!? This is a recent change that definitely should be reversed (along with the loss of the ability to right-click on menu headings to get a brief help box) cheers Chris On 19/07/2010 18:51, Steve Simon, P.Mean Consulting wrote: > There was a question that I just stumbled across a month late, but I > wanted to bring it up again, not because the earlier answers were bad, > but rather because it illustrates a philosophical change in how SPSS > handles data analysis. I don't like this change and I wanted to see what > others think about it. > > ANDRES ALBERTO BURGA LEON wrote: > >> I need to check for differences among four independent samples that made >> a short essay (rated using a 3 point rating scale= good, medium, poor). >> I presume I could use Kruskal-Wallis H or a similar non-parametric test, >> but in the new PASW 18 non-parametric test “Test Field” option only >> accepts variable measured in scale level. Can’t I use ordinal variables? >> What test could I do? > > I just got SPSS 18 loaded last week, and I looked at the non-parametric > test procedure. They've integrated all the nonparametric tests into a > single menu choice, which may or may not be a good thing, but > interestingly, this appears to be the first (but probably not the last) > statistical procedure that uses the nominal/ordinal/scale properties of > the variable. The graphical methods, of course, have been using this > nominal/ordinal/scale property for quite a while (since version 15, I > believe). > > This is something like the philosophy implemented in JMP, but it is, so > far, only implemented partially in SPSS. I'm not sure I like this new > approach and I thought it would be worth discussing this on this list. > > In theory, the measurement property of a particular variable should > allow you to use or eliminate certain tests or graphs, but there are two > problems with this. First, a lot of times, you want to run a test that > doesn't quite meet the measurement properties of the variable in > question just as a sensitivity check. Second, SPSS does a lousy job of > assigning measurement properties to a file that is imported from another > source. I dislike the idea of having to check and fix the measurement > properties of every variable in every imported data set. > > The advantage is that incorporating measurement properties into all the > statistical procedures might prevent an inexperienced data analyst from > making a bogus choice and might end up steering them towards a more > appropriate choice. It also is a potential time saver in that you will > be presented with a smaller number of valid choices for your graphs and > analyses. > > The workaround is to change the measurement properties temporarily, but > I find this tedious and annoying. > > Another workaround is to use the legacy dialogs, which did not have > these restrictions. I'm not thrilled about this either. I don't want to > teach people to use something that is obsolete. > > I also see this as a common question that I will get in consulting (why > doesn't SPSS let me run the X procedure on my data set). It might end up > padding my consulting income, but I still don't like it. > > What do other people think about this use of measurement properties as a > gatekeeper that prevents certain graphs/analyses from being run? > > Steve Simon, Standard Disclaimer > "Data entry and data management issues with examples > in IBM SPSS," Tuesday, August 24, 11am-noon CDT. > Free webinar. Details at www.pmean.com/webinars > > ===================== > 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: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independent samples)
|
|
Bear in mind that you can easily set the measurement level you want either temporarily or permanently in either syntax or most dialog boxes (right click in the source list), although not in the new nonparametric dialog, unfortunately. In fact, some Statistics dialog boxes have had measurement-level sensitivity as far back as version 11.5. In Custom Tables, you can even use the same variable with multiple measurement levels in the same table. Our R extension dialogs automatically map categorical MLs to R factors. And the Define Variable Properties dialog has a set of heuristics to help users set an appropriate level - not perfect, of course, but often a good start. So this is meant as a convenience and guide, not as a prescription. Regards Jon Peck SPSS, an IBM Company [hidden email] 312-651-3435
I also agree strongly. There are times that variables that might be nominal in one situation may actually be interval/ratio in another context (e.g., Lord's treatise "On the Statistical Treatment of Football Numbers" from the early 1950s). The Steven's view of measurement is not universally accepted, with a number of psychometricians and statisticians arguing against the NOIR categorization. (I can post some of these references if anyone is interested.) At bottom, my sense is that the software should not limit the researcher in any sort of mechanically enforced fashion. Those of us who know what we are doing should not have to go through the extra hoops that such a system imposes. My two cents' worth . . . Harley Dr. Harley Baker Professor and Chair, Psychology Program Sage Hall 2061 California State University Channel Islands One University Drive Camarillo, CA 93012 805.437.8997 (p) 805.437.8951 (f) [hidden email] ________________________________________ From: SPSSX(r) Discussion [[hidden email]] on behalf of Dr C B Stride [[hidden email]] Sent: Monday, July 19, 2010 12:04 PM To: [hidden email] Subject: Re: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples) I strongly agree with you Steve; and the argument for this system, i.e. that it will stop users running tests using variables of the wrong type is weak in itself in that if users don't understand why their test is appropriate they are unlikely to be able to interpret the results properly anyway... I can also think of instances where using a method not normally associated with certain types of vars can be useful e.g. mean tables on a dichotomous variable will give you a proportion scoring 1. And if these restrictions contuinue to be implemented, will we be stopped from adding dummy variables to regression!? This is a recent change that definitely should be reversed (along with the loss of the ability to right-click on menu headings to get a brief help box) cheers Chris On 19/07/2010 18:51, Steve Simon, P.Mean Consulting wrote: > There was a question that I just stumbled across a month late, but I > wanted to bring it up again, not because the earlier answers were bad, > but rather because it illustrates a philosophical change in how SPSS > handles data analysis. I don't like this change and I wanted to see what > others think about it. > > ANDRES ALBERTO BURGA LEON wrote: > >> I need to check for differences among four independent samples that made >> a short essay (rated using a 3 point rating scale= good, medium, poor). >> I presume I could use Kruskal-Wallis H or a similar non-parametric test, >> but in the new PASW 18 non-parametric test “Test Field” option only >> accepts variable measured in scale level. Can’t I use ordinal variables? >> What test could I do? > > I just got SPSS 18 loaded last week, and I looked at the non-parametric > test procedure. They've integrated all the nonparametric tests into a > single menu choice, which may or may not be a good thing, but > interestingly, this appears to be the first (but probably not the last) > statistical procedure that uses the nominal/ordinal/scale properties of > the variable. The graphical methods, of course, have been using this > nominal/ordinal/scale property for quite a while (since version 15, I > believe). > > This is something like the philosophy implemented in JMP, but it is, so > far, only implemented partially in SPSS. I'm not sure I like this new > approach and I thought it would be worth discussing this on this list. > > In theory, the measurement property of a particular variable should > allow you to use or eliminate certain tests or graphs, but there are two > problems with this. First, a lot of times, you want to run a test that > doesn't quite meet the measurement properties of the variable in > question just as a sensitivity check. Second, SPSS does a lousy job of > assigning measurement properties to a file that is imported from another > source. I dislike the idea of having to check and fix the measurement > properties of every variable in every imported data set. > > The advantage is that incorporating measurement properties into all the > statistical procedures might prevent an inexperienced data analyst from > making a bogus choice and might end up steering them towards a more > appropriate choice. It also is a potential time saver in that you will > be presented with a smaller number of valid choices for your graphs and > analyses. > > The workaround is to change the measurement properties temporarily, but > I find this tedious and annoying. > > Another workaround is to use the legacy dialogs, which did not have > these restrictions. I'm not thrilled about this either. I don't want to > teach people to use something that is obsolete. > > I also see this as a common question that I will get in consulting (why > doesn't SPSS let me run the X procedure on my data set). It might end up > padding my consulting income, but I still don't like it. > > What do other people think about this use of measurement properties as a > gatekeeper that prevents certain graphs/analyses from being run? > > Steve Simon, Standard Disclaimer > "Data entry and data management issues with examples > in IBM SPSS," Tuesday, August 24, 11am-noon CDT. > Free webinar. Details at www.pmean.com/webinars > > ===================== > 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: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples)
|
|
In reply to this post by Baker, Harley
There is nothing preventing you from changing the measurement level for any variable at any time (with the notable exception of treating a string variable as scale/continuous). So the software is not really imposing any "limit" on what you can do in that sense. For some procedures, measurement level affects the computation of the results; so there has to be some mechanism for identifying measurement level.
I also agree strongly. There are times that variables that might be nominal in one situation may actually be interval/ratio in another context (e.g., Lord's treatise "On the Statistical Treatment of Football Numbers" from the early 1950s). The Steven's view of measurement is not universally accepted, with a number of psychometricians and statisticians arguing against the NOIR categorization. (I can post some of these references if anyone is interested.) At bottom, my sense is that the software should not limit the researcher in any sort of mechanically enforced fashion. Those of us who know what we are doing should not have to go through the extra hoops that such a system imposes. My two cents' worth . . . Harley Dr. Harley Baker Professor and Chair, Psychology Program Sage Hall 2061 California State University Channel Islands One University Drive Camarillo, CA 93012 805.437.8997 (p) 805.437.8951 (f) [hidden email] ________________________________________ From: SPSSX(r) Discussion [[hidden email]] on behalf of Dr C B Stride [[hidden email]] Sent: Monday, July 19, 2010 12:04 PM To: [hidden email] Subject: Re: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples) I strongly agree with you Steve; and the argument for this system, i.e. that it will stop users running tests using variables of the wrong type is weak in itself in that if users don't understand why their test is appropriate they are unlikely to be able to interpret the results properly anyway... I can also think of instances where using a method not normally associated with certain types of vars can be useful e.g. mean tables on a dichotomous variable will give you a proportion scoring 1. And if these restrictions contuinue to be implemented, will we be stopped from adding dummy variables to regression!? This is a recent change that definitely should be reversed (along with the loss of the ability to right-click on menu headings to get a brief help box) cheers Chris On 19/07/2010 18:51, Steve Simon, P.Mean Consulting wrote: > There was a question that I just stumbled across a month late, but I > wanted to bring it up again, not because the earlier answers were bad, > but rather because it illustrates a philosophical change in how SPSS > handles data analysis. I don't like this change and I wanted to see what > others think about it. > > ANDRES ALBERTO BURGA LEON wrote: > >> I need to check for differences among four independent samples that made >> a short essay (rated using a 3 point rating scale= good, medium, poor). >> I presume I could use Kruskal-Wallis H or a similar non-parametric test, >> but in the new PASW 18 non-parametric test “Test Field” option only >> accepts variable measured in scale level. Can’t I use ordinal variables? >> What test could I do? > > I just got SPSS 18 loaded last week, and I looked at the non-parametric > test procedure. They've integrated all the nonparametric tests into a > single menu choice, which may or may not be a good thing, but > interestingly, this appears to be the first (but probably not the last) > statistical procedure that uses the nominal/ordinal/scale properties of > the variable. The graphical methods, of course, have been using this > nominal/ordinal/scale property for quite a while (since version 15, I > believe). > > This is something like the philosophy implemented in JMP, but it is, so > far, only implemented partially in SPSS. I'm not sure I like this new > approach and I thought it would be worth discussing this on this list. > > In theory, the measurement property of a particular variable should > allow you to use or eliminate certain tests or graphs, but there are two > problems with this. First, a lot of times, you want to run a test that > doesn't quite meet the measurement properties of the variable in > question just as a sensitivity check. Second, SPSS does a lousy job of > assigning measurement properties to a file that is imported from another > source. I dislike the idea of having to check and fix the measurement > properties of every variable in every imported data set. > > The advantage is that incorporating measurement properties into all the > statistical procedures might prevent an inexperienced data analyst from > making a bogus choice and might end up steering them towards a more > appropriate choice. It also is a potential time saver in that you will > be presented with a smaller number of valid choices for your graphs and > analyses. > > The workaround is to change the measurement properties temporarily, but > I find this tedious and annoying. > > Another workaround is to use the legacy dialogs, which did not have > these restrictions. I'm not thrilled about this either. I don't want to > teach people to use something that is obsolete. > > I also see this as a common question that I will get in consulting (why > doesn't SPSS let me run the X procedure on my data set). It might end up > padding my consulting income, but I still don't like it. > > What do other people think about this use of measurement properties as a > gatekeeper that prevents certain graphs/analyses from being run? > > Steve Simon, Standard Disclaimer > "Data entry and data management issues with examples > in IBM SPSS," Tuesday, August 24, 11am-noon CDT. > Free webinar. Details at www.pmean.com/webinars > > ===================== > 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: A Question about skewness
|
|
In reply to this post by stace swayne
stace swayne wrote:
> Dear List, > > Is there a rule of thumb about how much skewness is acceptable vs. unacceptable. > I have read conflicting opinions that say that skewness should be less than 2, > and then some people have said that skewness should be less than 1. > > > Hi Stace: If you are talking about the skewness coefficient provided by SPSS with DESCRIPTIVES or EXAMINE, the ratio of the coefficient by its standard error is a Z test that can be considered significant if its absolute value is greater than 1.96. But, since statistical significance and statistical relevance is not the same, take into account that for very big samples the result can be significant even if the coefficient is very low. Somewhere I read that a skewness coefficient over 1 (in absolute value) is important (I'm not talking of "significant"). HTH, Marta GG (we are the champions...) -- For miscellaneous SPSS related statistical stuff, visit: http://gjyp.nl/marta/ ===================== 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: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples)
|
In reply to this post by Rick Oliver-3
Dear all,
"For some procedures, measurement level affects the computation of the results; so there has to be some mechanism for identifying measurement level." >>>Agree. But just as with, say, CATREG, I think this mechanism -if really needed- should reside in the syntax or dialogue boxes. If my CATREG results suggest I can treat some variable as scale -within the scope of this single procedure- I want to modify this in just this single procedure, not in the data. Otherwise, in a next procedure, I may have to modify it back. Of course nothing prevents us from changing measurement levels of all numeric variables. For me the main issue would be that doing so increases the amount of work I need in order to get stuff done. This will become especially annoying when one wants to run procedures on one variable that would require different measurement levels. So I perfectly agree with previous opinions: if a data analyst doesn't know what he's doing, enforcing measurement levels upon certain procedures will probably not prevent him from producing 'less than optimal' results. Or as one colleague wisely phrased: "nothing can ever beat the stupidity of clients". So let's not sacrifice our educated users in a futile attempt to protect some less educated users from themselves. Another argument is that measurement levels are often disputable. More precisely, many variables that are strictly ordinal (like 5 point Likert scales) tend to be treated as scale variables in the social sciences. So if I decide to run chisq tests on those and someone else prefers t tests, I'll have to change all measurement levels with extra, unnecessary lines of syntax. As a compromise, perhaps, can't users be enabled to switch measurement level sensitivity on or off, like SET MEASUREMENTLEVELSENSITIVITY=OFF. Best, Ruben van den Berg Consultant Models & Methods TNS NIPO Email: [hidden email] Mobiel: +31 6 24641435 Telefoon: +31 20 522 5738 Internet: www.tns-nipo.com Date: Mon, 19 Jul 2010 15:17:32 -0500 From: [hidden email] Subject: Re: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples) To: [hidden email] There is nothing preventing you from changing the measurement level for any variable at any time (with the notable exception of treating a string variable as scale/continuous). So the software is not really imposing any "limit" on what you can do in that sense. For some procedures, measurement level affects the computation of the results; so there has to be some mechanism for identifying measurement level.
I also agree strongly. There are times that variables that might be nominal in one situation may actually be interval/ratio in another context (e.g., Lord's treatise "On the Statistical Treatment of Football Numbers" from the early 1950s). The Steven's view of measurement is not universally accepted, with a number of psychometricians and statisticians arguing against the NOIR categorization. (I can post some of these references if anyone is interested.) At bottom, my sense is that the software should not limit the researcher in any sort of mechanically enforced fashion. Those of us who know what we are doing should not have to go through the extra hoops that such a system imposes. My two cents' worth . . . Harley Dr. Harley Baker Professor and Chair, Psychology Program Sage Hall 2061 California State University Channel Islands One University Drive Camarillo, CA 93012 805.437.8997 (p) 805.437.8951 (f) [hidden email] ________________________________________ From: SPSSX(r) Discussion [[hidden email]] on behalf of Dr C B Stride [[hidden email]] Sent: Monday, July 19, 2010 12:04 PM To: [hidden email] Subject: Re: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples) I strongly agree with you Steve; and the argument for this system, i.e. that it will stop users running tests using variables of the wrong type is weak in itself in that if users don't understand why their test is appropriate they are unlikely to be able to interpret the results properly anyway... I can also think of instances where using a method not normally associated with certain types of vars can be useful e.g. mean tables on a dichotomous variable will give you a proportion scoring 1. And if these restrictions contuinue to be implemented, will we be stopped from adding dummy variables to regression!? This is a recent change that definitely should be reversed (along with the loss of the ability to right-click on menu headings to get a brief help box) cheers Chris On 19/07/2010 18:51, Steve Simon, P.Mean Consulting wrote: > There was a question that I just stumbled across a month late, but I > wanted to bring it up again, not because the earlier answers were bad, > but rather because it illustrates a philosophical change in how SPSS > handles data analysis. I don't like this change and I wanted to see what > others think about it. > > ANDRES ALBERTO BURGA LEON wrote: > >> I need to check for differences among four independent samples that made >> a short essay (rated using a 3 point rating scale= good, medium, poor). >> I presume I could use Kruskal-Wallis H or a similar non-parametric test, >> but in the new PASW 18 non-parametric test “Test Field” option only >> accepts variable measured in scale level. Can’t I use ordinal variables? >> What test could I do? > > I just got SPSS 18 loaded last week, and I looked at the non-parametric > test procedure. They've integrated all the nonparametric tests into a > single menu choice, which may or may not be a good thing, but > interestingly, this appears to be the first (but probably not the last) > statistical procedure that uses the nominal/ordinal/scale properties of > the variable. The graphical methods, of course, have been using this > nominal/ordinal/scale property for quite a while (since version 15, I > believe). > > This is something like the philosophy implemented in JMP, but it is, so > far, only implemented partially in SPSS. I'm not sure I like this new > approach and I thought it would be worth discussing this on this list. > > In theory, the measurement property of a particular variable should > allow you to use or eliminate certain tests or graphs, but there are two > problems with this. First, a lot of times, you want to run a test that > doesn't quite meet the measurement properties of the variable in > question just as a sensitivity check. Second, SPSS does a lousy job of > assigning measurement properties to a file that is imported from another > source. I dislike the idea of having to check and fix the measurement > properties of every variable in every imported data set. > > The advantage is that incorporating measurement properties into all the > statistical procedures might prevent an inexperienced data analyst from > making a bogus choice and might end up steering them towards a more > appropriate choice. It also is a potential time saver in that you will > be presented with a smaller number of valid choices for your graphs and > analyses. > > The workaround is to change the measurement properties temporarily, but > I find this tedious and annoying. > > Another workaround is to use the legacy dialogs, which did not have > these restrictions. I'm not thrilled about this either. I don't want to > teach people to use something that is obsolete. > > I also see this as a common question that I will get in consulting (why > doesn't SPSS let me run the X procedure on my data set). It might end up > padding my consulting income, but I still don't like it. > > What do other people think about this use of measurement properties as a > gatekeeper that prevents certain graphs/analyses from being run? > > Steve Simon, Standard Disclaimer > "Data entry and data management issues with examples > in IBM SPSS," Tuesday, August 24, 11am-noon CDT. > Free webinar. Details at www.pmean.com/webinars > > ===================== > 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 New Windows 7: Find the right PC for you. Learn more. |
Re: SPSS is becoming too rigid (Was: Comparing ordinal variable in k Independant samples)
|
|
Dear Listers, I agree with the other contributors. Enforcing measurement
level restrictions makes things more unwieldy, and the benefits are debatable. Inexperienced users should be encouraged to consult tutorials,
examples and textbooks to develop their understanding. Being forced down
certain routes by SPSS – without understanding why – may even
add to a beginner’s confusion. As others have pointed out, measurement levels are something of
a moveable feast, and I don’t want to end up with a situation where
repeating a procedure on a number of different variables requires that they are
all assigned the same measurement level. Garry Gelade Business Analytic Ltd From: SPSSX(r) Discussion
[mailto:[hidden email]] On Behalf Of Ruben van den Berg Dear
all, Date: Mon, 19 Jul 2010 15:17:32 -0500
New
Windows 7: Find the right PC for you. Learn more. |
| Free forum by Nabble | Edit this page |