Curious about K-S test in NPTESTS
Locked
Curious about K-S test in NPTESTS
 
|
Hi all, I am working to compare various sample distributions and thought that the Kolmogorov-Smirnov test in NPTESTS would be appropriate. However, it doesn’t reveal
these two obviously different distributions as significant. Syntax is below. Any insight would be appreciated. Jason DATA LIST FREE/FORM X FREQ. BEGIN DATA. 1 1 7 1 2 24 1 3 38 1 4 24 1 5 7 2 1 20 2 2 20 2 3 20 2 4 20 2 5 20 END DATA. WEIGHT BY FREQ. VARIABLE LEVEL X (SCALE). *Nonparametric Tests: Independent Samples. NPTESTS /INDEPENDENT TEST (X) GROUP (FORM) KOLMOGOROV_SMIRNOV /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE /CRITERIA ALPHA=0.05 CILEVEL=95. _____________________________________________________________ Jason W. Beckstead, Ph.D. Associate Professor/Quantitative Methodologist University of South Florida College of Public Health 13201 Bruce B. Downs Blvd., MDC56, Tampa, FL 33612, USA Statistical Editor, International Journal of Nursing Studies phone: +1.813.974.7667 fax: +1.813.974.4719 personal website: http://personal.health.usf.edu/jbeckste/ International Journal of Nursing Studies http://www.elsevier.com/ijns |
|
|
If you duplicate the data, so that you have 20 records, the sig level goes from .367 to .068. Triple it and the sig is .013. Even using Exact Tests (on the original data), the K-S test sig level is .146. Even the Mann-Whitney test does not detect the difference. So you need more data. On Sat, Dec 8, 2018 at 2:22 PM Beckstead, Jason <[hidden email]> wrote:
|
Re: Curious about K-S test in NPTESTS
|
|
In reply to this post by Beckstead, Jason
Your apparent difference is not location (mean), but "scale"
(standard deviation). Since the K-S test is sensitive to both, it
will not be especially powerful for either.
A second generic problem is that rank-based tests, in derivation
and in most casual discussion, assume that there are no ties -- but
your data features only 5 values for 200 cases. Many ties. Sometimes
having ties is mainly notable for making the test-values wrong (so,
Jon's use of "exact stats" for p-level is good practice); sometimes,
like here, it also is reflected in a weakened test.
I put the 10 cells in a 2x5 contingency table, ignoring order, and the
online calculator gave me X2= 18.83, p< .001 (4 d.f.), which looks okay.
--
Rich Ulrich
From: SPSSX(r) Discussion <[hidden email]> on behalf of Beckstead, Jason <[hidden email]>
Sent: Saturday, December 8, 2018 4:21 PM To: [hidden email] Subject: Curious about K-S test in NPTESTS Hi all, I am working to compare various sample distributions and thought that the Kolmogorov-Smirnov test in NPTESTS would be appropriate. However, it doesn’t reveal these two obviously different distributions as significant. Syntax is below. Any insight would be appreciated. Jason
DATA LIST FREE/FORM X FREQ. BEGIN DATA. 1 1 7 1 2 24 1 3 38 1 4 24 1 5 7 2 1 20 2 2 20 2 3 20 2 4 20 2 5 20 END DATA. WEIGHT BY FREQ. VARIABLE LEVEL X (SCALE). *Nonparametric Tests: Independent Samples. NPTESTS /INDEPENDENT TEST (X) GROUP (FORM) KOLMOGOROV_SMIRNOV /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE /CRITERIA ALPHA=0.05 CILEVEL=95.
_____________________________________________________________ Jason W. Beckstead, Ph.D. Associate Professor/Quantitative Methodologist University of South Florida College of Public Health 13201 Bruce B. Downs Blvd., MDC56, Tampa, FL 33612, USA Statistical Editor, International Journal of Nursing Studies phone: +1.813.974.7667 fax: +1.813.974.4719 personal website: http://personal.health.usf.edu/jbeckste/ International Journal of Nursing Studies http://www.elsevier.com/ijns
|
Re: Curious about K-S test in NPTESTS
|
|
Thanks to all. You guys confirmed my hunch. I guess that the best way to compare the difference between two or more distributions of 5 values (like those resulting from different sets of verbal anchors on Likert-type scales) is to use chi-square test.
Jason
sent via me phone
From: Rich Ulrich <[hidden email]>
Sent: Saturday, December 8, 2018 8:19:14 PM To: [hidden email]; Beckstead, Jason Subject: Re: Curious about K-S test in NPTESTS
Your apparent difference is not location (mean), but "scale"
(standard deviation). Since the K-S test is sensitive to both, it
will not be especially powerful for either.
A second generic problem is that rank-based tests, in derivation
and in most casual discussion, assume that there are no ties -- but
your data features only 5 values for 200 cases. Many ties. Sometimes
having ties is mainly notable for making the test-values wrong (so,
Jon's use of "exact stats" for p-level is good practice); sometimes,
like here, it also is reflected in a weakened test.
I put the 10 cells in a 2x5 contingency table, ignoring order, and the
online calculator gave me X2= 18.83, p< .001 (4 d.f.), which looks okay.
--
Rich Ulrich
From: SPSSX(r) Discussion <[hidden email]> on behalf of Beckstead, Jason <[hidden email]>
Sent: Saturday, December 8, 2018 4:21 PM To: [hidden email] Subject: Curious about K-S test in NPTESTS Hi all, I am working to compare various sample distributions and thought that the Kolmogorov-Smirnov test in NPTESTS would be appropriate. However, it doesn’t reveal these two obviously different distributions as significant. Syntax is below. Any insight would be appreciated. Jason
DATA LIST FREE/FORM X FREQ. BEGIN DATA. 1 1 7 1 2 24 1 3 38 1 4 24 1 5 7 2 1 20 2 2 20 2 3 20 2 4 20 2 5 20 END DATA. WEIGHT BY FREQ. VARIABLE LEVEL X (SCALE). *Nonparametric Tests: Independent Samples. NPTESTS /INDEPENDENT TEST (X) GROUP (FORM) KOLMOGOROV_SMIRNOV /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE /CRITERIA ALPHA=0.05 CILEVEL=95.
_____________________________________________________________ Jason W. Beckstead, Ph.D. Associate Professor/Quantitative Methodologist University of South Florida College of Public Health 13201 Bruce B. Downs Blvd., MDC56, Tampa, FL 33612, USA Statistical Editor, International Journal of Nursing Studies phone: +1.813.974.7667 fax: +1.813.974.4719 personal website: http://personal.health.usf.edu/jbeckste/ International Journal of Nursing Studies http://www.elsevier.com/ijns
|
Re: Curious about K-S test in NPTESTS
|
|
A likert-type scaling has order, and the ordinary contingency table
does not respect that. You have a 2 d.f. test rather than a 4 d.f.
test, and the /usual/ difference between distributions is the mean.
Having a hole in likert-type responses (1s, least number of 2s, 3s,
4s, 5s) would be unusual and suspicious all by itself, but is among
what the contingency table tests.
If you want a test on variances, in addition to the usual test on means,
the "powerful" manner of testing is to conduct two separate tests, each
with 1 d.f. I think it also would be clearer in a report on results to describe
two tests.
For the simple variance-ratio F-test, I get F=1.923, p < .001 (same p-report
I gave for the 2x5 table). With more precision: this one is p= 0.00069, while
the 2x5 table was p= 0.0008-something.
--
Rich Ulrich
From: Beckstead, Jason <[hidden email]>
Sent: Sunday, December 9, 2018 1:14 PM To: [hidden email]; Rich Ulrich Subject: Re: Curious about K-S test in NPTESTS
Thanks to all. You guys confirmed my hunch. I guess that the best way to compare the difference between two or more distributions of 5 values (like those resulting from different sets of verbal anchors on Likert-type scales) is to use chi-square test.
Jason
sent via me phone
From: Rich Ulrich <[hidden email]>
Sent: Saturday, December 8, 2018 8:19:14 PM To: [hidden email]; Beckstead, Jason Subject: Re: Curious about K-S test in NPTESTS
Your apparent difference is not location (mean), but "scale"
(standard deviation). Since the K-S test is sensitive to both, it
will not be especially powerful for either.
A second generic problem is that rank-based tests, in derivation
and in most casual discussion, assume that there are no ties -- but
your data features only 5 values for 200 cases. Many ties. Sometimes
having ties is mainly notable for making the test-values wrong (so,
Jon's use of "exact stats" for p-level is good practice); sometimes,
like here, it also is reflected in a weakened test.
I put the 10 cells in a 2x5 contingency table, ignoring order, and the
online calculator gave me X2= 18.83, p< .001 (4 d.f.), which looks okay.
--
Rich Ulrich
From: SPSSX(r) Discussion <[hidden email]> on behalf of Beckstead, Jason <[hidden email]>
Sent: Saturday, December 8, 2018 4:21 PM To: [hidden email] Subject: Curious about K-S test in NPTESTS Hi all, I am working to compare various sample distributions and thought that the Kolmogorov-Smirnov test in NPTESTS would be appropriate. However, it doesn’t reveal these two obviously different distributions as significant. Syntax is below. Any insight would be appreciated. Jason
DATA LIST FREE/FORM X FREQ. BEGIN DATA. 1 1 7 1 2 24 1 3 38 1 4 24 1 5 7 2 1 20 2 2 20 2 3 20 2 4 20 2 5 20 END DATA. WEIGHT BY FREQ. VARIABLE LEVEL X (SCALE). *Nonparametric Tests: Independent Samples. NPTESTS /INDEPENDENT TEST (X) GROUP (FORM) KOLMOGOROV_SMIRNOV /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE /CRITERIA ALPHA=0.05 CILEVEL=95.
_____________________________________________________________ Jason W. Beckstead, Ph.D. Associate Professor/Quantitative Methodologist University of South Florida College of Public Health 13201 Bruce B. Downs Blvd., MDC56, Tampa, FL 33612, USA Statistical Editor, International Journal of Nursing Studies phone: +1.813.974.7667 fax: +1.813.974.4719 personal website: http://personal.health.usf.edu/jbeckste/ International Journal of Nursing Studies http://www.elsevier.com/ijns
|
Re: Curious about K-S test in NPTESTS
Administrator
|
CROSSTABS also reports a Chi-square test of linear-by-linear association,
which you could use in this situation. See David Howell's nice note on it here: https://www.uvm.edu/~dhowell/StatPages/More_Stuff/OrdinalChisq/OrdinalChiSq.html By the way, for those who may not have heard, David Howell died on Oct 4 of this year. He'd been fighting cancer for about 8 years. Rich Ulrich wrote > A likert-type scaling has order, and the ordinary contingency table > does not respect that. You have a 2 d.f. test rather than a 4 d.f. > test, and the /usual/ difference between distributions is the mean. > Having a hole in likert-type responses (1s, least number of 2s, 3s, > 4s, 5s) would be unusual and suspicious all by itself, but is among > what the contingency table tests. > > If you want a test on variances, in addition to the usual test on means, > the "powerful" manner of testing is to conduct two separate tests, each > with 1 d.f. I think it also would be clearer in a report on results to > describe > two tests. > > For the simple variance-ratio F-test, I get F=1.923, p < .001 (same > p-report > I gave for the 2x5 table). With more precision: this one is p= 0.00069, > while > the 2x5 table was p= 0.0008-something. > > -- > Rich Ulrich > > ________________________________ > From: Beckstead, Jason < > jbeckste@.usf > > > Sent: Sunday, December 9, 2018 1:14 PM > To: > spssx-l@.uga > ; Rich Ulrich > Subject: Re: Curious about K-S test in NPTESTS > > Thanks to all. You guys confirmed my hunch. I guess that the best way to > compare the difference between two or more distributions of 5 values (like > those resulting from different sets of verbal anchors on Likert-type > scales) is to use chi-square test. > Jason > > sent via me phone > > ________________________________ > From: Rich Ulrich < > rich-ulrich@ > > > Sent: Saturday, December 8, 2018 8:19:14 PM > To: > SPSSX-L@.UGA > ; Beckstead, Jason > Subject: Re: Curious about K-S test in NPTESTS > > Your apparent difference is not location (mean), but "scale" > (standard deviation). Since the K-S test is sensitive to both, it > will not be especially powerful for either. > > A second generic problem is that rank-based tests, in derivation > and in most casual discussion, assume that there are no ties -- but > your data features only 5 values for 200 cases. Many ties. Sometimes > having ties is mainly notable for making the test-values wrong (so, > Jon's use of "exact stats" for p-level is good practice); sometimes, > like here, it also is reflected in a weakened test. > > I put the 10 cells in a 2x5 contingency table, ignoring order, and the > online calculator gave me X2= 18.83, p< .001 (4 d.f.), which looks okay. > > -- > Rich Ulrich > > > > ________________________________ > From: SPSSX(r) Discussion < > SPSSX-L@.UGA > > on behalf of Beckstead, Jason < > jbeckste@.usf > > > Sent: Saturday, December 8, 2018 4:21 PM > To: > SPSSX-L@.UGA > Subject: Curious about K-S test in NPTESTS > > > Hi all, > > I am working to compare various sample distributions and thought that the > Kolmogorov-Smirnov test in NPTESTS would be appropriate. However, it > doesn’t reveal these two obviously different distributions as significant. > Syntax is below. Any insight would be appreciated. > > Jason > > > > DATA LIST FREE/FORM X FREQ. > > BEGIN DATA. > > 1 1 7 > > 1 2 24 > > 1 3 38 > > 1 4 24 > > 1 5 7 > > 2 1 20 > > 2 2 20 > > 2 3 20 > > 2 4 20 > > 2 5 20 > > END DATA. > > WEIGHT BY FREQ. > > VARIABLE LEVEL X (SCALE). > > *Nonparametric Tests: Independent Samples. > > NPTESTS > > /INDEPENDENT TEST (X) GROUP (FORM) KOLMOGOROV_SMIRNOV > > /MISSING SCOPE=ANALYSIS USERMISSING=EXCLUDE > > /CRITERIA ALPHA=0.05 CILEVEL=95. > > > > _____________________________________________________________ > > Jason W. Beckstead, Ph.D. > > Associate Professor/Quantitative Methodologist > > University of South Florida College of Public Health > > 13201 Bruce B. Downs Blvd., MDC56, Tampa, FL 33612, USA > > Statistical Editor, International Journal of Nursing Studies > > phone: +1.813.974.7667 fax: +1.813.974.4719 > > personal website: http://personal.health.usf.edu/jbeckste/ > > International Journal of Nursing Studies http://www.elsevier.com/ijns > > > > ===================== To manage your subscription to SPSSX-L, send a > message to > LISTSERV@.UGA > <mailto: > LISTSERV@.UGA > > (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 > LISTSERV@.UGA > (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 ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.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
--
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: Curious about K-S test in NPTESTS
|
|
Bruce,
You need a period in that URL: https://www.uvm.edu/~dhowell/StatPages/More_Stuff/OrdinalChisq/OrdinalChiSq.html Sad news about David Howell. Tony Babinec -----Original Message----- From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Bruce Weaver Sent: Sunday, December 9, 2018 4:13 PM To: [hidden email] Subject: Re: Curious about K-S test in NPTESTS CROSSTABS also reports a Chi-square test of linear-by-linear association, which you could use in this situation. See David Howell's nice note on it here: https://www.uvm.edu/~dhowell/StatPages/More_Stuff/OrdinalChisq/OrdinalChiSqhtml By the way, for those who may not have heard, David Howell died on Oct 4 of this year. He'd been fighting cancer for about 8 years. Rich Ulrich wrote > A likert-type scaling has order, and the ordinary contingency table > does not respect that. You have a 2 d.f. test rather than a 4 d.f. > test, and the /usual/ difference between distributions is the mean. > Having a hole in likert-type responses (1s, least number of 2s, 3s, > 4s, 5s) would be unusual and suspicious all by itself, but is among > what the contingency table tests. > > If you want a test on variances, in addition to the usual test on > means, the "powerful" manner of testing is to conduct two separate tests, each > with 1 d.f. I think it also would be clearer in a report on results to > describe > two tests. > > For the simple variance-ratio F-test, I get F=1.923, p < .001 (same > p-report I gave for the 2x5 table). With more precision: this one is > p= 0.00069, while the 2x5 table was p= 0.0008-something. > > -- > Rich Ulrich > > ________________________________ ===================== 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: Curious about K-S test in NPTESTS
|
|
I don't know why that particular comment from David Howell's page was
included, but the test and p-value for non-linear component described the
data on Howell's page.
Applying the algorithm for "non-linear" to the artificial data from Jason -
the linear component is exactly 0, so the achieved X2 value is also the
3-df test, with smaller p-value because it has 3 df, not 4.
Sad, about David.
--
Rich Ulrich
From: SPSSX(r) Discussion <[hidden email]> on behalf of Anthony Babinec <[hidden email]>
Sent: Sunday, December 9, 2018 5:34 PM To: [hidden email] Subject: Re: Curious about K-S test in NPTESTS Bruce,
You need a period in that URL: https://www.uvm.edu/~dhowell/StatPages/More_Stuff/OrdinalChisq/OrdinalChiSq.html Sad news about David Howell. Tony Babinec -----Original Message----- From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Bruce Weaver Sent: Sunday, December 9, 2018 4:13 PM To: [hidden email] Subject: Re: Curious about K-S test in NPTESTS CROSSTABS also reports a Chi-square test of linear-by-linear association, which you could use in this situation. See David Howell's nice note on it here: https://www.uvm.edu/~dhowell/StatPages/More_Stuff/OrdinalChisq/OrdinalChiSqhtml By the way, for those who may not have heard, David Howell died on Oct 4 of this year. He'd been fighting cancer for about 8 years. Rich Ulrich wrote > A likert-type scaling has order, and the ordinary contingency table > does not respect that. You have a 2 d.f. test rather than a 4 d.f. > test, and the /usual/ difference between distributions is the mean. > Having a hole in likert-type responses (1s, least number of 2s, 3s, > 4s, 5s) would be unusual and suspicious all by itself, but is among > what the contingency table tests. > > If you want a test on variances, in addition to the usual test on > means, the "powerful" manner of testing is to conduct two separate tests, each > with 1 d.f. I think it also would be clearer in a report on results to > describe > two tests. > > For the simple variance-ratio F-test, I get F=1.923, p < .001 (same > p-report I gave for the 2x5 table). With more precision: this one is > p= 0.00069, while the 2x5 table was p= 0.0008-something. > > -- > Rich Ulrich > > ________________________________ ===================== 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: Curious about K-S test in NPTESTS
|
Administrator
|
Rich is right. Here are the results of Howell's analysis on the data posted
in this thread. Test Value df p Percent Pearson Chi-Square 18.832 4 .001 100.0 Linear-by-Linear Association .000 1 1.000 .0 Deviation from Linearity 18.832 3 .000 100.0 Rich Ulrich wrote > I don't know why that particular comment from David Howell's page was > included, but the test and p-value for non-linear component described the > data on Howell's page. > > Applying the algorithm for "non-linear" to the artificial data from Jason > - > the linear component is exactly 0, so the achieved X2 value is also the > 3-df test, with smaller p-value because it has 3 df, not 4. > > Sad, about David. > > -- > Rich Ulrich ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.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
--
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: Curious about K-S test in NPTESTS
|
Administrator
|
This post was updated on .
In reply to this post by Anthony Babinec
That's interesting. When I look at my post in Nabble (first link below), the
period is there. When I look at Tony's reply in Nabble (second link below), the period is missing. Gremlins, perhaps? ;-) http://spssx-discussion.1045642.n5.nabble.com/Curious-about-K-S-test-in-NPTESTS-tp5737122p5737127.html http://spssx-discussion.1045642.n5.nabble.com/Curious-about-K-S-test-in-NPTESTS-tp5737122p5737128.html EDIT: I mean that when I look at the ~copy~ of my post shown in Tony's reply, the period is missing! Anthony Babinec wrote > Bruce, > You need a period in that URL: > > https://www.uvm.edu/~dhowell/StatPages/More_Stuff/OrdinalChisq/OrdinalChiSq.html > > Sad news about David Howell. > > Tony Babinec ----- -- Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to LISTSERV@LISTSERV.UGA.EDU (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
--
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: Curious about K-S test in NPTESTS
|
|
In reply to this post by Bruce Weaver
Thanks to all of you who have weighed in on this one.
BIG PICTURE: I am investigating how the use of different sets of verbal anchors for 5-point Likert-type response scales alter the distributions of the responses to items. I have 6 different sets of labels. I have 27 items that were used to create 6 versions of a questionnaire. Each questionnaire contained all 27 items but used a different set of verbal anchors for the 5-point scale. Students from a very large class were randomly assigned to complete one of the questionnaires. I have approximately 125 subjects in each of the 6 conditions. Analysis plan: 1) for each item I am comparing the distribution of responses across conditions using a chi-square test. I am not using the linear-by-linear test, because it is my understanding that this test assumes that the distances between adjacent categories is equal. 2) also, I am conducting item-metric repeated measures ANOVAs. In this approach, the items are treated as the subjects (n=27). For each item I have computed standard deviation, skewness, and kurtosis. One ANOVA, for example, treats the standard deviations of the items a the DV. I repeat this type of analysis on the item-skewnesses, and on the kurtosis values. Any comments, or suggestions welcome. Jason _____________________________________________________________ Jason W. Beckstead, Ph.D. Associate Professor/Quantitative Methodologist University of South Florida College of Public Health 13201 Bruce B. Downs Blvd., MDC56, Tampa, FL 33612, USA Statistical Editor, International Journal of Nursing Studies phone: +1.813.974.7667 fax: +1.813.974.4719 personal website: http://personal.health.usf.edu/jbeckste/ International Journal of Nursing Studies http://www.elsevier.com/ijns -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Bruce Weaver Sent: Monday, December 10, 2018 8:11 AM To: [hidden email] Subject: Re: Curious about K-S test in NPTESTS Rich is right. Here are the results of Howell's analysis on the data posted in this thread. Test Value df p Percent Pearson Chi-Square 18.832 4 .001 100.0 Linear-by-Linear Association .000 1 1.000 .0 Deviation from Linearity 18.832 3 .000 100.0 Rich Ulrich wrote > I don't know why that particular comment from David Howell's page was > included, but the test and p-value for non-linear component described the > data on Howell's page. > > Applying the algorithm for "non-linear" to the artificial data from Jason > - > the linear component is exactly 0, so the achieved X2 value is also the > 3-df test, with smaller p-value because it has 3 df, not 4. > > Sad, about David. > > -- > Rich Ulrich ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.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 |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |