Mc Nemar Test for 1:2 Case-Control ?
Locked
12
12
Mc Nemar Test for 1:2 Case-Control ?
 
|
Dear Members, Consider the 1:2 matched study where each case is matched with 2 controls. Health status (classified as "+" or "-" ) of all cases and controls were recorded. Summary of the gathered data looks like as follows: Group HealthStatus Count Case + 50 Case - 50 Control + 10 Control - 190 Can we use McNemar Test to test if the proportion of cases on the "+" HealthStatus is greater than the controls? Please suggest an appropriate test if McNemar test is not appropriate. Thank you. Eins |
Re: Mc Nemar Test for 1:2 Case-Control ?
|
|
Hi:
Look for stratified Mantel-Haenszel odds ratio (available at CROSSTABS), or use conditional logistic regression (SPSS can be tricked to do it using a Cox regression model). HTH, Marta GG El 10/03/2016 a las 6:46, E. Bernardo
escribió:
===================== 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: Mc Nemar Test for 1:2 Case-Control ?
|
|
In reply to this post by E. Bernardo
Ooooops! Do you have only summarized data or raw data?
You will not be able to permor any of the two methods I proposed before unless your data set looks like this: MatchGroup CaseControl HealthStatus 1 1 1 1 0 0 1 0 Whatever... 2 1 0 2 0 1 2 0 Whatever... 3 ... 3 ... 3 4 4 4 ... El 10/03/2016 a las 6:46, E. Bernardo
escribió:
===================== 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: Mc Nemar Test for 1:2 Case-Control ?
|
|
In reply to this post by E. Bernardo
As Marta G-G says, you are very limited if you don't have the raw data.
=====================
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
And, abstractly, I'm not grasping how the McNemar Test would fit, even for a simple pair. Given the summary numbers, you have a simple contingency table. If those are the numbers, there is not much doubt about a difference. However, if the "matching" was important, then taking it into account would reduce the size of the apparent effect. That's simple logic. ("Why do we want to match? - Because those variables might account for the outcome.") Often, trying to match is useful for collecting/selecting Controls to use. However, I have long been hostile to analyzing such data as "matched", given (a) the imprecision of much of such matching ("age within 4 years"); (b) the loss of d.f. in the analysis; and, (c) the weakness of treating the matching covariates essentially as categories instead of as continuous measures. - The alternative of using the matching variables as covariates (in a logistic regression, here) is almost bound to be more powerful and more robust. - That, in turn, requires that the data be entered with a line for each subject, with the Group, Status, and personal covariates. -- Rich Ulrich Date: Thu, 10 Mar 2016 05:46:45 +0000 From: [hidden email] Subject: Mc Nemar Test for 1:2 Case-Control ? To: [hidden email] Dear Members, Consider the 1:2 matched study where each case is matched with 2 controls. Health status (classified as "+" or "-" ) of all cases and controls were recorded. Summary of the gathered data looks like as follows: Group HealthStatus Count Case + 50 Case - 50 Control + 10 Control - 190 Can we use McNemar Test to test if the proportion of cases on the "+" HealthStatus is greater than the controls? Please suggest an appropriate test if McNemar test is not appropriate. Thank you. |
Re: Mc Nemar Test for 1:2 Case-Control ?
|
|
Thank you for your comments, Marta and Rich. Yes, raw data are available. The summary data are just my example to illustrate my data. Actually, I tried to use the C regression (as trick to conduct Conditional regression in SPSS). However, the SPSS reports only the -2LL. I dont know if I did the correct trick. Marta, following the variable names you provided in the previous email, here is my syntax: DATASET ACTIVATE DataSet2. COXREG Casecontrol /STATUS=HeathStatus /STRATA=Matchgroup /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). What is wrong with my syntax? Thank you. On Friday, March 11, 2016 2:27 AM, Rich Ulrich <[hidden email]> wrote: As Marta G-G says, you are very limited if you don't have the raw data. And, abstractly, I'm not grasping how the McNemar Test would fit, even for a simple pair. Given the summary numbers, you have a simple contingency table. If those are the numbers, there is not much doubt about a difference. However, if the "matching" was important, then taking it into account would reduce the size of the apparent effect. That's simple logic. ("Why do we want to match? - Because those variables might account for the outcome.") Often, trying to match is useful for collecting/selecting Controls to use. However, I have long been hostile to analyzing such data as "matched", given (a) the imprecision of much of such matching ("age within 4 years"); (b) the loss of d.f. in the analysis; and, (c) the weakness of treating the matching covariates essentially as categories instead of as continuous measures. - The alternative of using the matching variables as covariates (in a logistic regression, here) is almost bound to be more powerful and more robust. - That, in turn, requires that the data be entered with a line for each subject, with the Group, Status, and personal covariates. -- Rich Ulrich Date: Thu, 10 Mar 2016 05:46:45 +0000 From: [hidden email] Subject: Mc Nemar Test for 1:2 Case-Control ? To: [hidden email] Dear Members, Consider the 1:2 matched study where each case is matched with 2 controls. Health status (classified as "+" or "-" ) of all cases and controls were recorded. Summary of the gathered data looks like as follows: Group HealthStatus Count Case + 50 Case - 50 Control + 10 Control - 190 Can we use McNemar Test to test if the proportion of cases on the "+" HealthStatus is greater than the controls? Please suggest an appropriate test if McNemar test is not appropriate. Thank you. =====================
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: Mc Nemar Test for 1:2 Case-Control ?
|
|
Hi
I don't think your syntax is correct. You have a create a false " time variable" first , let's call it pseudotime, with values EQ 2 for controls and EQ 1 for cases. This variable acts as time on the COXREG command. Status is defined by CaseControl, the covariable/factor is HealthStatus, and Matchgroup acts as strata variable. Did you try the Mantel-Haenszel statistic approach? Sorry I can't be more detailed/helpful, but I'm at an airport right now, with just my laptop (away from the desktop office computer with all my old syntax collection. Regards, Marta Garcia-Granero El 12/03/2016 a las 4:36, E. Bernardo
escribió:
===================== 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: Mc Nemar Test for 1:2 Case-Control ?
|
|
I remembered Raynald had the code I posted some years ago on his
great webpage:
http://spsstools.net/en/syntax/447/ HTH, Marta El 12/03/2016 a las 10:35, Marta
Garcia-Granero escribió:
Hi ===================== 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: Mc Nemar Test for 1:2 Case-Control ?
|
|
Whilst acknowledging Marta's contributions, you might like to post an enquiry on the Google Group MedStats...I'm biased, I founded it :). There are all the most notable statisticians (eg Bland, Altman, Senn, Harrell Jnr, Campbell to name but a few.) To post you need to join; thereafter all discussions are visible. Best Wishes, Martin P. Holt Freelance Medical Statistician If you can't explain it simply, you don't understand it well enough.....Einstein Linked In: https://www.linkedin.com/in/martin-holt-3b800b48?trk=nav_responsive_tab_profile
|
Re: Mc Nemar Test for 1:2 Case-Control ?
Administrator
|
As Martin knows, I am also a member of the MedStats group. It is indeed a very good group for discussion of medical statistics (or biostatistics). However, based on what I have seen there, I would not recommend it as a place to seek advice on how to do a given task using SPSS. The likely response would be something like, <dismissive tone>"Why are you using SPSS?</dismissive tone> You ought to be using R/SAS/Stata!"
Perhaps I'm overstating a bit...but just a bit! ;-) IMHO, this forum will be pretty hard to beat when it comes to advice on how to do task X using SPSS. 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: Mc Nemar Test for 1:2 Case-Control ?
|
|
In reply to this post by Marta Garcia-Granero
Thank you for the link. I think my SPSS syntax now is correct. COXREG faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). However there is warning as part of the outputs. This warning is: "Since coefficients did not converge, no further models will be fitted." And the odds ratio is too large: Exp(B) = 74.769 Is the odds ratio wrong? SPSS outputs are pasted below. Eins
On Saturday, March 12, 2016 5:35 PM, Marta Garcia-Granero <[hidden email]> wrote:
Hi I don't think your syntax is correct. You have a create a false " time variable" first , let's call it pseudotime, with values EQ 2 for controls and EQ 1 for cases. This variable acts as time on the COXREG command. Status is defined by CaseControl, the covariable/factor is HealthStatus, and Matchgroup acts as strata variable. Did you try the Mantel-Haenszel statistic approach? Sorry I can't be more detailed/helpful, but I'm at an airport right now, with just my laptop (away from the desktop office computer with all my old syntax collection. Regards, Marta Garcia-Granero El 12/03/2016 a las 4:36, E. Bernardo
escribió:
=====================
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: Mc Nemar Test for 1:2 Case-Control ?
|
Administrator
|
Did you try increasing the number of iterations on the /CRITERIA sub-command? E.g.,
COXREG faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X /CRITERIA=PIN(.05) POUT(.10) ITERATE(50).
--
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: Mc Nemar Test for 1:2 Case-Control ?
|
|
Thank you Bruce for your comments. The warning was "Since coefficients did not converge, no further models will be fitted." The Iteration History states that "At least one coefficient is tending to infinity after 3 iterations". So increasing the number of iteration from 20 to 50 does not help since it stopped at the 3rd iteration. Any further suggestion to solve the problem is highly solicited. Thank you. On Wednesday, March 23, 2016 2:04 AM, Bruce Weaver <[hidden email]> wrote: Did you try increasing the number of iterations on the /CRITERIA sub-command? E.g., COXREG faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X /CRITERIA=PIN(.05) POUT(.10) ITERATE(50). E. Bernardo wrote > Thank you for the link. I think my SPSS syntax now is correct.COXREG > faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X > /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). > However there is warning as part of the outputs. This warning is:"Since > coefficients did not converge, no further models will be fitted." > And the odds ratio is too large: Exp(B) = 74.769 > Is the odds ratio wrong? > SPSS outputs are pasted below. > Eins > > > | Warnings | | | | | | | | | | > | Since coefficients did not converge, no further models will be fitted. | > | | | | | | | | | > | | | | | | | | | | | > | Case Processing Summary | | | | | | | > | | N | Percent | | | | | | | > | Cases available in analysis | Eventa | 47 | 33.3% | | | | | | | > | Censored | 94 | 66.7% | | | | | | | > | Total | 141 | 100.0% | | | | | | | > | Cases dropped | Cases with missing values | 0 | 0.0% | | | | | | | > | Cases with negative time | 0 | 0.0% | | | | | | | > | Censored cases before the earliest event in a stratum | 0 | 0.0% | | | > | | | | > | Total | 0 | 0.0% | | | | | | | > | Total | 141 | 100.0% | | | | | | | > | a. Dependent Variable: faketime | | | | | | | > | | | | | | | | | | | > | Stratum Statusa | | | | | | | > | Stratum | Event | Censored | Censored Percent | | | | | | | > | 1.0 | 1 | 2 | 66.7% | | | | | | | > | 2.0 | 1 | 2 | 66.7% | | | | | | | > | 3.0 | 1 | 2 | 66.7% | | | | | | | > | 4.0 | 1 | 2 | 66.7% | | | | | | | > | 5.0 | 1 | 2 | 66.7% | | | | | | | > | 6.0 | 1 | 2 | 66.7% | | | | | | | > | 7.0 | 1 | 2 | 66.7% | | | | | | | > | 8.0 | 1 | 2 | 66.7% | | | | | | | > | 9.0 | 1 | 2 | 66.7% | | | | | | | > | 10.0 | 1 | 2 | 66.7% | | | | | | | > | 11.0 | 1 | 2 | 66.7% | | | | | | | > | 12.0 | 1 | 2 | 66.7% | | | | | | | > | 13.0 | 1 | 2 | 66.7% | | | | | | | > | 14.0 | 1 | 2 | 66.7% | | | | | | | > | 15.0 | 1 | 2 | 66.7% | | | | | | | > | 16.0 | 1 | 2 | 66.7% | | | | | | | > | 17.0 | 1 | 2 | 66.7% | | | | | | | > | 18.0 | 1 | 2 | 66.7% | | | | | | | > | 19.0 | 1 | 2 | 66.7% | | | | | | | > | 20.0 | 1 | 2 | 66.7% | | | | | | | > | 21.0 | 1 | 2 | 66.7% | | | | | | | > | 22.0 | 1 | 2 | 66.7% | | | | | | | > | 23.0 | 1 | 2 | 66.7% | | | | | | | > | 24.0 | 1 | 2 | 66.7% | | | | | | | > | 25.0 | 1 | 2 | 66.7% | | | | | | | > | 26.0 | 1 | 2 | 66.7% | | | | | | | > | 27.0 | 1 | 2 | 66.7% | | | | | | | > | 28.0 | 1 | 2 | 66.7% | | | | | | | > | 29.0 | 1 | 2 | 66.7% | | | | | | | > | 30.0 | 1 | 2 | 66.7% | | | | | | | > | 31.0 | 1 | 2 | 66.7% | | | | | | | > | 32.0 | 1 | 2 | 66.7% | | | | | | | > | 33.0 | 1 | 2 | 66.7% | | | | | | | > | 34.0 | 1 | 2 | 66.7% | | | | | | | > | 35.0 | 1 | 2 | 66.7% | | | | | | | > | 36.0 | 1 | 2 | 66.7% | | | | | | | > | 37.0 | 1 | 2 | 66.7% | | | | | | | > | 38.0 | 1 | 2 | 66.7% | | | | | | | > | 39.0 | 1 | 2 | 66.7% | | | | | | | > | 40.0 | 1 | 2 | 66.7% | | | | | | | > | 41.0 | 1 | 2 | 66.7% | | | | | | | > | 42.0 | 1 | 2 | 66.7% | | | | | | | > | 43.0 | 1 | 2 | 66.7% | | | | | | | > | 44.0 | 1 | 2 | 66.7% | | | | | | | > | 45.0 | 1 | 2 | 66.7% | | | | | | | > | 46.0 | 1 | 2 | 66.7% | | | | | | | > | 47.0 | 1 | 2 | 66.7% | | | | | | | > | Total | 47 | 94 | 66.7% | | | | | | | > | a. The strata variable is : Stratum | | | | | | | > | | | | | | | | | | | > | | | | | | | | | | | > | Block 0: Beginning Block | | | | | | | | | | > | | | | | | | | | | | > | Omnibus Tests of Model Coefficients | | | | | | | | | | > | -2 Log Likelihood | | | | | | | | | | > | 103.270 | | | | | | | | | | > | | | | | | | | | | | > | | | | | | | | | | | > | Block 1: Method = Enter | | | | | | | | | | > | | | | | | | | | | | > | Iteration Historyb | | | | | | | | > | | -2 Log Likelihooda | Coefficient | | | | | | | | > | X | | | | | | | | > | 1 | 80.334 | 2.071 | | | | | | | | > | 2 | 76.835 | 3.255 | | | | | | | | > | 3 | 75.745 | 4.314 | | | | | | | | > | a. Beginning Block Number 0, initial Log Likelihood function: -2 Log > likelihood: 103.270 | | | | | | | | > | b. At least one coefficient is tending to infinity after 3 iterations | > | | | | | | | > | | | | | | | | | | | > | Omnibus Tests of Model Coefficientsa | > | -2 Log Likelihood | Overall (score) | Change From Previous Step | Change > From Previous Block | > | Chi-square | df | Sig. | Chi-square | df | Sig. | Chi-square | df | Sig. > | > | 75.745 | 20.024 | 1 | .000 | 27.524 | 1 | .000 | 27.524 | 1 | .000 | > | a. Beginning Block Number 1. Method = Enter | > | | | | | | | | | | | > | Variables in the Equation | | | | > | | B | SE | Wald | df | Sig. | Exp(B) | | | | > | X | 4.314 | 1.861 | 5.375 | 1 | .020 | 74.769 | | | | > | | | | | | | | | | | > | Covariate Means | | | | | | | | | > | | Mean | | | | | | | | | > | X | .262 | | | | | | | | | > > > > On Saturday, March 12, 2016 5:35 PM, Marta Garcia-Granero < > mgarciagranero@ > > wrote: > > > Hi > > I don't think your syntax is correct. You have a create a false " time > variable" first , let's call it pseudotime, with values EQ 2 for controls > and EQ 1 for cases. This variable acts as time on the COXREG command. > Status is defined by CaseControl, the covariable/factor is HealthStatus, > and Matchgroup acts as strata variable. > > Did you try the Mantel-Haenszel statistic approach? > > Sorry I can't be more detailed/helpful, but I'm at an airport right now, > with just my laptop (away from the desktop office computer with all my old > syntax collection. > > Regards, > Marta Garcia-Granero > > El 12/03/2016 a las 4:36, E. Bernardo escribió: > > Thank you for your comments, Marta and Rich. > Yes, raw data are available. The summary data are just my example to > illustrate my data. Actually, I tried to use the C regression (as trick to > conduct Conditional regression in SPSS). However, the SPSS reports only > the -2LL. I dont know if I did the correct trick. > Marta, following the variable names you provided in the previous email, > here is my syntax: > DATASET ACTIVATE DataSet2. COXREG Casecontrol /STATUS=HeathStatus > /STRATA=Matchgroup /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). > What is wrong with my syntax? > Thank you. > > On Friday, March 11, 2016 2:27 AM, Rich Ulrich < > rich-ulrich@ > > wrote: > > > #yiv5634543243 --.yiv5634543243hmmessage > P{margin:0px;padding:0px;}#yiv5634543243 > body.yiv5634543243hmmessage{font-size:12pt;font-family:Calibri;}#yiv5634543243 > As Marta G-G says, you are very limited if you don't have the raw data. > And, abstractly, I'm not grasping how the McNemar Test would fit, even > for a > simple pair. > > Given the summary numbers, you have a simple contingency table. > If those are the numbers, there is not much doubt about a difference. > > However, if the "matching" was important, then taking it into account > would reduce the size of the apparent effect. That's simple logic. ("Why > do we > want to match? - Because those variables might account for the > outcome.") > > Often, trying to match is useful for collecting/selecting Controls to > use. However, > I have long been hostile to analyzing such data as "matched", given (a) > the > imprecision of much of such matching ("age within 4 years"); (b) the loss > of > d.f. in the analysis; and, (c) the weakness of treating the matching > covariates > essentially as categories instead of as continuous measures. - The > alternative > of using the matching variables as covariates (in a logistic regression, > here) > is almost bound to be more powerful and more robust. - That, in turn, > requires > that the data be entered with a line for each subject, with the Group, > Status, > and personal covariates. > > -- > Rich Ulrich > > > > Date: Thu, 10 Mar 2016 05:46:45 +0000 > From: > [hidden email] > Subject: Mc Nemar Test for 1:2 Case-Control ? > To: > [hidden email] > > Dear Members, > Consider the 1:2 matched study where each case is matched with 2 > controls. Health status (classified as "+" or "-" ) of all cases and > controls were recorded. Summary of the gathered data looks like as > follows: > Group HealthStatus Count Case + 50 > Case - 50 Control + > 10 Control - 190 > Can we use McNemar Test to test if the proportion of cases on the "+" > HealthStatus is greater than the controls? Please suggest an appropriate > test if McNemar test is not appropriate. > Thank you. > ===================== 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 > [hidden email] > (not to SPSSX-L), with no body text except thecommand. To leave the list, > send the commandSIGNOFF SPSSX-LFor a list of commands to manage > subscriptions, send the commandINFO 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 ----- -- 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. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Mc-Nemar-Test-for-1-2-Case-Control-tp5731696p5731793.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 |
Re: Mc Nemar Test for 1:2 Case-Control ?
|
|
Your original data showed an OR of 19, with 300 cases -
=====================
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
The 2x2 table had cells of {50,50; 190;10}. Is 78 unreasonable? This run shows N= 141. Rather a loss. What does the 2x2 table look like? (And, What happened to the other half of the data?) Getting an infinite coefficient is a symptom of having perfect separation. I never used CoxReg for case control, so I can't vouch for your syntax - in case "perfect separation" is not the answer. -- Rich Ulrich Date: Wed, 23 Mar 2016 01:25:39 +0000 From: [hidden email] Subject: Re: Mc Nemar Test for 1:2 Case-Control ? To: [hidden email] Thank you Bruce for your comments. The warning was "Since coefficients did not converge, no further models will be fitted." The Iteration History states that "At least one coefficient is tending to infinity after 3 iterations". So increasing the number of iteration from 20 to 50 does not help since it stopped at the 3rd iteration. Any further suggestion to solve the problem is highly solicited. Thank you. On Wednesday, March 23, 2016 2:04 AM, Bruce Weaver <[hidden email]> wrote: Did you try increasing the number of iterations on the /CRITERIA sub-command? E.g., COXREG faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X /CRITERIA=PIN(.05) POUT(.10) ITERATE(50). E. Bernardo wrote > Thank you for the link. I think my SPSS syntax now is correct.COXREG > faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X > /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). > However there is warning as part of the outputs. This warning is:"Since > coefficients did not converge, no further models will be fitted." > And the odds ratio is too large: Exp(B) = 74.769 > Is the odds ratio wrong? > SPSS outputs are pasted below. > Eins > > > | Warnings | | | | | | | | | | > | Since coefficients did not converge, no further models will be fitted. | > | | | | | | | | | > | | | | | | | | | | | > | Case Processing Summary | | | | | | | > | | N | Percent | | | | | | | > | Cases available in analysis | Eventa | 47 | 33.3% | | | | | | | > | Censored | 94 | 66.7% | | | | | | | > | Total | 141 | 100.0% | | | | | | | > | Cases dropped | Cases with missing values | 0 | 0.0% | | | | | | | > | Cases with negative time | 0 | 0.0% | | | | | | | > | Censored cases before the earliest event in a stratum | 0 | 0.0% | | | > | | | | > | Total | 0 | 0.0% | | | | | | | > | Total | 141 | 100.0% | | | | | | | |
Re: Mc Nemar Test for 1:2 Case-Control ?
|
|
Hi Rich and others, It is a case-control study using 1 case and 2 controls. There are 47 cases and 94 controls. The 2 x2 Data subjected to Conditional Logistic Regression is shown in the table below. Results: OR = 74.769 and 95% CI of OR: (1.949 - 2868.900). Any comment on the outputs?
On Wednesday, March 23, 2016 12:16 PM, Rich Ulrich <[hidden email]> wrote: Your original data showed an OR of 19, with 300 cases - The 2x2 table had cells of {50,50; 190;10}. Is 78 unreasonable? This run shows N= 141. Rather a loss. What does the 2x2 table look like? (And, What happened to the other half of the data?) Getting an infinite coefficient is a symptom of having perfect separation. I never used CoxReg for case control, so I can't vouch for your syntax - in case "perfect separation" is not the answer. -- Rich Ulrich Date: Wed, 23 Mar 2016 01:25:39 +0000 From: [hidden email] Subject: Re: Mc Nemar Test for 1:2 Case-Control ? To: [hidden email] Thank you Bruce for your comments. The warning was "Since coefficients did not converge, no further models will be fitted." The Iteration History states that "At least one coefficient is tending to infinity after 3 iterations". So increasing the number of iteration from 20 to 50 does not help since it stopped at the 3rd iteration. Any further suggestion to solve the problem is highly solicited. Thank you. On Wednesday, March 23, 2016 2:04 AM, Bruce Weaver <[hidden email]> wrote: Did you try increasing the number of iterations on the /CRITERIA sub-command? E.g., COXREG faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X /CRITERIA=PIN(.05) POUT(.10) ITERATE(50). E. Bernardo wrote > Thank you for the link. I think my SPSS syntax now is correct.COXREG > faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X > /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). > However there is warning as part of the outputs. This warning is:"Since > coefficients did not converge, no further models will be fitted." > And the odds ratio is too large: Exp(B) = 74.769 > Is the odds ratio wrong? > SPSS outputs are pasted below. > Eins > > > | Warnings | | | | | | | | | | > | Since coefficients did not converge, no further models will be fitted. | > | | | | | | | | | > | | | | | | | | | | | > | Case Processing Summary | | | | | | | > | | N | Percent | | | | | | | > | Cases available in analysis | Eventa | 47 | 33.3% | | | | | | | > | Censored | 94 | 66.7% | | | | | | | > | Total | 141 | 100.0% | | | | | | | > | Cases dropped | Cases with missing values | 0 | 0.0% | | | | | | | > | Cases with negative time | 0 | 0.0% | | | | | | | > | Censored cases before the earliest event in a stratum | 0 | 0.0% | | | > | | | | > | Total | 0 | 0.0% | | | | | | | > | Total | 141 | 100.0% | | | | | | | =====================
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: Mc Nemar Test for 1:2 Case-Control ?
|
Administrator
|
In reply to this post by E. Bernardo
Okay! Have you tried GENLIN using GEE as an alternative to conditional logistic regression? Several years ago, I estimated some conditional logistic regression models using Stata, then some time later, analyzed the same data using GENLIN with GEE. The results were very similar. You can see the comparisons in this old thread from comp.soft-sys.stat.spss:
https://groups.google.com/forum/#!topic/comp.soft-sys.stat.spss/zkxR016mZxM If I understand your COXREG command, I think the GENLIN syntax would be something like this: * Generalized Estimating Equations. GENLIN outcome (REFERENCE=FIRST) WITH X /MODEL X INTERCEPT=YES DISTRIBUTION=BINOMIAL LINK=LOGIT /REPEATED SUBJECT=Stratum SORT=YES CORRTYPE=UNSTRUCTURED ADJUSTCORR=YES COVB=ROBUST MAXITERATIONS=100 PCONVERGE=1e-006(ABSOLUTE) UPDATECORR=1 /MISSING CLASSMISSING=EXCLUDE /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION. This assumes X is continuous. If it is categorical, change WITH to BY on the first line. This model may not converge either, which would suggest there's some problem with your data. But I think it's worth a try. HTH.
--
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: Mc Nemar Test for 1:2 Case-Control ?
|
|
Dear Bruce, I modified the first line of your syntax by changing WITH to BY because X is categorical with two values (0 and 1). I also put (order=descending) as shown below. * Generalized Estimating Equations. GENLIN outcome (REFERENCE=FIRST) BY X (Order = descending) /MODEL X INTERCEPT=YES DISTRIBUTION=BINOMIAL LINK=LOGIT /REPEATED SUBJECT=Stratum SORT=YES CORRTYPE=UNSTRUCTURED ADJUSTCORR=YES COVB=ROBUST MAXITERATIONS=100 PCONVERGE=1e-006(ABSOLUTE) UPDATECORR=1 /MISSING CLASSMISSING=EXCLUDE /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION. /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION. The GEE outputs are as follows:
How will be compute the Odds Ratio? Is it OR = Exp(B) = exp(.000066959532) = 1.000067? The B coefficients look weird! Any comment about the outputs? Thank you. Eins On Thursday, March 24, 2016 12:24 AM, Bruce Weaver <[hidden email]> wrote: Okay! Have you tried GENLIN using GEE as an alternative to conditional logistic regression? Several years ago, I estimated some conditional logistic regression models using Stata, then some time later, analyzed the same data using GENLIN with GEE. The results were very similar. You can see the comparisons in this old thread from comp.soft-sys.stat.spss: https://groups.google.com/forum/#!topic/comp.soft-sys.stat.spss/zkxR016mZxM If I understand your COXREG command, I think the GENLIN syntax would be something like this: * Generalized Estimating Equations. GENLIN outcome (REFERENCE=FIRST) WITH X /MODEL X INTERCEPT=YES DISTRIBUTION=BINOMIAL LINK=LOGIT /REPEATED SUBJECT=Stratum SORT=YES CORRTYPE=UNSTRUCTURED ADJUSTCORR=YES COVB=ROBUST MAXITERATIONS=100 PCONVERGE=1e-006(ABSOLUTE) UPDATECORR=1 /MISSING CLASSMISSING=EXCLUDE /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION. This assumes X is continuous. If it is categorical, change WITH to BY on the first line. This model may not converge either, which would suggest there's some problem with your data. But I think it's worth a try. HTH. E. Bernardo wrote > Thank you Bruce for your comments. > The warning was "Since coefficients did not converge, no further models > will be fitted."The Iteration History states that "At least one > coefficient is tending to infinity after 3 iterations".So increasing the > number of iteration from 20 to 50 does not help since it stopped at the > 3rd iteration. Any further suggestion to solve the problem is highly > solicited. > > Thank you. > > > On Wednesday, March 23, 2016 2:04 AM, Bruce Weaver < > bruce.weaver@ > > wrote: > > > Did you try increasing the number of iterations on the /CRITERIA > sub-command? > E.g., > > COXREG faketime > /STATUS=outcome(1) > /STRATA=Stratum > /METHOD=ENTER X > /CRITERIA=PIN(.05) POUT(.10) ITERATE(50). > > > > > E. Bernardo wrote >> Thank you for the link. I think my SPSS syntax now is correct.COXREG >> faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X >> /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). >> However there is warning as part of the outputs. This warning is:"Since >> coefficients did not converge, no further models will be fitted." >> And the odds ratio is too large: Exp(B) = 74.769 >> Is the odds ratio wrong? >> SPSS outputs are pasted below. >> Eins >> >> >> | Warnings | | | | | | | | | | >> | Since coefficients did not converge, no further models will be fitted. >> | >> | | | | | | | | | >> | | | | | | | | | | | >> | Case Processing Summary | | | | | | | >> | | N | Percent | | | | | | | >> | Cases available in analysis | Eventa | 47 | 33.3% | | | | | | | >> | Censored | 94 | 66.7% | | | | | | | >> | Total | 141 | 100.0% | | | | | | | >> | Cases dropped | Cases with missing values | 0 | 0.0% | | | | | | >> | >> | Cases with negative time | 0 | 0.0% | | | | | | | >> | Censored cases before the earliest event in a stratum | 0 | 0.0% | | >> | >> | | | | >> | Total | 0 | 0.0% | | | | | | | >> | Total | 141 | 100.0% | | | | | | | >> | a. Dependent Variable: faketime | | | | | | | >> | | | | | | | | | | | >> | Stratum Statusa | | | | | | | >> | Stratum | Event | Censored | Censored Percent | | | | | | | >> | 1.0 | 1 | 2 | 66.7% | | | | | | | >> | 2.0 | 1 | 2 | 66.7% | | | | | | | >> | 3.0 | 1 | 2 | 66.7% | | | | | | | >> | 4.0 | 1 | 2 | 66.7% | | | | | | | >> | 5.0 | 1 | 2 | 66.7% | | | | | | | >> | 6.0 | 1 | 2 | 66.7% | | | | | | | >> | 7.0 | 1 | 2 | 66.7% | | | | | | | >> | 8.0 | 1 | 2 | 66.7% | | | | | | | >> | 9.0 | 1 | 2 | 66.7% | | | | | | | >> | 10.0 | 1 | 2 | 66.7% | | | | | | | >> | 11.0 | 1 | 2 | 66.7% | | | | | | | >> | 12.0 | 1 | 2 | 66.7% | | | | | | | >> | 13.0 | 1 | 2 | 66.7% | | | | | | | >> | 14.0 | 1 | 2 | 66.7% | | | | | | | >> | 15.0 | 1 | 2 | 66.7% | | | | | | | >> | 16.0 | 1 | 2 | 66.7% | | | | | | | >> | 17.0 | 1 | 2 | 66.7% | | | | | | | >> | 18.0 | 1 | 2 | 66.7% | | | | | | | >> | 19.0 | 1 | 2 | 66.7% | | | | | | | >> | 20.0 | 1 | 2 | 66.7% | | | | | | | >> | 21.0 | 1 | 2 | 66.7% | | | | | | | >> | 22.0 | 1 | 2 | 66.7% | | | | | | | >> | 23.0 | 1 | 2 | 66.7% | | | | | | | >> | 24.0 | 1 | 2 | 66.7% | | | | | | | >> | 25.0 | 1 | 2 | 66.7% | | | | | | | >> | 26.0 | 1 | 2 | 66.7% | | | | | | | >> | 27.0 | 1 | 2 | 66.7% | | | | | | | >> | 28.0 | 1 | 2 | 66.7% | | | | | | | >> | 29.0 | 1 | 2 | 66.7% | | | | | | | >> | 30.0 | 1 | 2 | 66.7% | | | | | | | >> | 31.0 | 1 | 2 | 66.7% | | | | | | | >> | 32.0 | 1 | 2 | 66.7% | | | | | | | >> | 33.0 | 1 | 2 | 66.7% | | | | | | | >> | 34.0 | 1 | 2 | 66.7% | | | | | | | >> | 35.0 | 1 | 2 | 66.7% | | | | | | | >> | 36.0 | 1 | 2 | 66.7% | | | | | | | >> | 37.0 | 1 | 2 | 66.7% | | | | | | | >> | 38.0 | 1 | 2 | 66.7% | | | | | | | >> | 39.0 | 1 | 2 | 66.7% | | | | | | | >> | 40.0 | 1 | 2 | 66.7% | | | | | | | >> | 41.0 | 1 | 2 | 66.7% | | | | | | | >> | 42.0 | 1 | 2 | 66.7% | | | | | | | >> | 43.0 | 1 | 2 | 66.7% | | | | | | | >> | 44.0 | 1 | 2 | 66.7% | | | | | | | >> | 45.0 | 1 | 2 | 66.7% | | | | | | | >> | 46.0 | 1 | 2 | 66.7% | | | | | | | >> | 47.0 | 1 | 2 | 66.7% | | | | | | | >> | Total | 47 | 94 | 66.7% | | | | | | | >> | a. The strata variable is : Stratum | | | | | | | >> | | | | | | | | | | | >> | | | | | | | | | | | >> | Block 0: Beginning Block | | | | | | | | | | >> | | | | | | | | | | | >> | Omnibus Tests of Model Coefficients | | | | | | | | | | >> | -2 Log Likelihood | | | | | | | | | | >> | 103.270 | | | | | | | | | | >> | | | | | | | | | | | >> | | | | | | | | | | | >> | Block 1: Method = Enter | | | | | | | | | | >> | | | | | | | | | | | >> | Iteration Historyb | | | | | | | | >> | | -2 Log Likelihooda | Coefficient | | | | | | | | >> | X | | | | | | | | >> | 1 | 80.334 | 2.071 | | | | | | | | >> | 2 | 76.835 | 3.255 | | | | | | | | >> | 3 | 75.745 | 4.314 | | | | | | | | >> | a. Beginning Block Number 0, initial Log Likelihood function: -2 Log >> likelihood: 103.270 | | | | | | | | >> | b. At least one coefficient is tending to infinity after 3 iterations | >> | | | | | | | >> | | | | | | | | | | | >> | Omnibus Tests of Model Coefficientsa | >> | -2 Log Likelihood | Overall (score) | Change From Previous Step | >> Change >> From Previous Block | >> | Chi-square | df | Sig. | Chi-square | df | Sig. | Chi-square | df | >> Sig. >> | >> | 75.745 | 20.024 | 1 | .000 | 27.524 | 1 | .000 | 27.524 | 1 | .000 | >> | a. Beginning Block Number 1. Method = Enter | >> | | | | | | | | | | | >> | Variables in the Equation | | | | >> | | B | SE | Wald | df | Sig. | Exp(B) | | | | >> | X | 4.314 | 1.861 | 5.375 | 1 | .020 | 74.769 | | | | >> | | | | | | | | | | | >> | Covariate Means | | | | | | | | | >> | | Mean | | | | | | | | | >> | X | .262 | | | | | | | | | >> >> >> >> On Saturday, March 12, 2016 5:35 PM, Marta Garcia-Granero < > >> mgarciagranero@ > >> > wrote: >> >> >> Hi >> >> I don't think your syntax is correct. You have a create a false " time >> variable" first , let's call it pseudotime, with values EQ 2 for controls >> and EQ 1 for cases. This variable acts as time on the COXREG command. >> Status is defined by CaseControl, the covariable/factor is HealthStatus, >> and Matchgroup acts as strata variable. >> >> Did you try the Mantel-Haenszel statistic approach? >> >> Sorry I can't be more detailed/helpful, but I'm at an airport right now, >> with just my laptop (away from the desktop office computer with all my >> old >> syntax collection. >> >> Regards, >> Marta Garcia-Granero >> >> El 12/03/2016 a las 4:36, E. Bernardo escribió: >> >> Thank you for your comments, Marta and Rich. >> Yes, raw data are available. The summary data are just my example to >> illustrate my data. Actually, I tried to use the C regression (as trick >> to >> conduct Conditional regression in SPSS). However, the SPSS reports only >> the -2LL. I dont know if I did the correct trick. >> Marta, following the variable names you provided in the previous email, >> here is my syntax: >> DATASET ACTIVATE DataSet2. COXREG Casecontrol /STATUS=HeathStatus >> /STRATA=Matchgroup /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). >> What is wrong with my syntax? >> Thank you. >> >> On Friday, March 11, 2016 2:27 AM, Rich Ulrich < > >> rich-ulrich@ > >> > wrote: >> >> >> #yiv5634543243 --.yiv5634543243hmmessage >> P{margin:0px;padding:0px;}#yiv5634543243 >> body.yiv5634543243hmmessage{font-size:12pt;font-family:Calibri;}#yiv5634543243 >> As Marta G-G says, you are very limited if you don't have the raw data. >> And, abstractly, I'm not grasping how the McNemar Test would fit, even >> for a >> simple pair. >> >> Given the summary numbers, you have a simple contingency table. >> If those are the numbers, there is not much doubt about a difference. >> >> However, if the "matching" was important, then taking it into account >> would reduce the size of the apparent effect. That's simple logic. >> do we >> want to match? - Because those variables might account for the >> outcome.") >> >> Often, trying to match is useful for collecting/selecting Controls to >> use. However, >> I have long been hostile to analyzing such data as "matched", given (a) >> the >> imprecision of much of such matching ("age within 4 years"); (b) the loss >> of >> d.f. in the analysis; and, (c) the weakness of treating the matching >> covariates >> essentially as categories instead of as continuous measures. - The >> alternative >> of using the matching variables as covariates (in a logistic regression, >> here) >> is almost bound to be more powerful and more robust. - That, in turn, >> requires >> that the data be entered with a line for each subject, with the Group, >> Status, >> and personal covariates. >> >> -- >> Rich Ulrich >> >> >> >> Date: Thu, 10 Mar 2016 05:46:45 +0000 >> From: > >> [hidden email] > >> Subject: Mc Nemar Test for 1:2 Case-Control ? >> To: > >> [hidden email] > >> >> Dear Members, >> Consider the 1:2 matched study where each case is matched with 2 >> controls. Health status (classified as "+" or "-" ) of all cases and >> controls were recorded. Summary of the gathered data looks like as >> follows: >> Group HealthStatus Count Case + 50 >> Case - 50 Control + >> 10 Control - 190 >> Can we use McNemar Test to test if the proportion of cases on the "+" >> HealthStatus is greater than the controls? Please suggest an appropriate >> test if McNemar test is not appropriate. >> Thank you. >> ===================== 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 > >> [hidden email] > >> (not to SPSSX-L), with no body text except thecommand. To leave the >> send the commandSIGNOFF SPSSX-LFor a list of commands to manage >> subscriptions, send the commandINFO 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 > > > > > > ----- > -- > Bruce Weaver > bweaver@ > 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. > > -- > View this message in context: > http://spssx-discussion.1045642.n5.nabble.com/Mc-Nemar-Test-for-1-2-Case-Control-tp5731696p5731793.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 ----- -- 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. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Mc-Nemar-Test-for-1-2-Case-Control-tp5731696p5731801.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 |
|||||||||||||||||||||||||||||||||||||||||||||
Re: Mc Nemar Test for 1:2 Case-Control ?
|
Administrator
|
Hi Eins. Yes, OR = Exp(B). If you add (EXPONENTIATED) after SOLUTION on the /PRINT sub-command, you'll see Exp(B) displayed in the table of coefficients (with a confidence interval).
An OR = 1.000067 normally would be taken to indicate no important effect. Despite being such a small effect size, the Wald test value is huge (Chi-square = 23.582, df = 1, p < .0001). What is the context? (If you gave it earlier in the thread, I missed it.) What is the outcome variable, and what is the dichotomous explanatory variable? 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: Mc Nemar Test for 1:2 Case-Control ?
|
|
Outcome = 1, 0 (1 case, 2 control) IV = 1, 0 (pain, no pain) On Thursday, March 31, 2016 2:40 AM, Bruce Weaver <[hidden email]> wrote: Hi Eins. Yes, OR = Exp(B). If you add (EXPONENTIATED) after SOLUTION on the /PRINT sub-command, you'll see Exp(B) displayed in the table of coefficients (with a confidence interval). An OR = 1.000067 normally would be taken to indicate no important effect. Despite being such a small effect size, the Wald test value is huge (Chi-square = 23.582, df = 1, p < .0001). What is the context? (If you gave it earlier in the thread, I missed it.) What is the outcome variable, and what is the dichotomous explanatory variable? Cheers, Bruce E. Bernardo-2 wrote > Dear Bruce, > I modified the first line of your syntax by changing WITH to BY because X > is categorical with two values (0 and 1). I also put (order=descending) > as shown below. > * Generalized Estimating Equations. > GENLIN outcome (REFERENCE=FIRST) BY X (Order = descending) > /MODEL X INTERCEPT=YES DISTRIBUTION=BINOMIAL LINK=LOGIT/REPEATED > SUBJECT=Stratum SORT=YES CORRTYPE=UNSTRUCTURED ADJUSTCORR=YES > COVB=ROBUST MAXITERATIONS=100 PCONVERGE=1e-006(ABSOLUTE) > UPDATECORR=1/MISSING CLASSMISSING=EXCLUDE/PRINT CPS DESCRIPTIVES MODELINFO > FIT SUMMARY SOLUTION./PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY > SOLUTION. > > The GEE outputs are as follows: > > | Parameter | B | Std. Error | 95% Wald Confidence Interval | Hypothesis > Test | > | Lower | Upper | Wald Chi-Square | df | Sig. | > | (Intercept) | -0.000046015394 | 0.000005938070 | -0.000057653798 | > -0.000034376990 | 60.050 | 1 | .000 | > | [X=1.00] | 0.000066959532 | 0.000013788672 | 0.000039934232 | > 0.000093984832 | 23.582 | 1 | .000 | > | [X=.00] | 0a | | | | | | | > | (Scale) | 1 | | | | | | | > > > How will be compute the Odds Ratio? Is it OR = Exp(B) = exp(.000066959532) > = 1.000067?The B coefficients look weird! Any comment about the outputs? > Thank you.Eins > > On Thursday, March 24, 2016 12:24 AM, Bruce Weaver < > bruce.weaver@ > > wrote: > > > Okay! Have you tried GENLIN using GEE as an alternative to conditional > logistic regression? Several years ago, I estimated some conditional > logistic regression models using Stata, then some time later, analyzed the > same data using GENLIN with GEE. The results were very similar. You can > see the comparisons in this old thread from comp.soft-sys.stat.spss: > > https://groups.google.com/forum/#!topic/comp.soft-sys.stat.spss/zkxR016mZxM > > If I understand your COXREG command, I think the GENLIN syntax would be > something like this: > > * Generalized Estimating Equations. > GENLIN outcome (REFERENCE=FIRST) WITH X > /MODEL X INTERCEPT=YES > DISTRIBUTION=BINOMIAL LINK=LOGIT > /REPEATED SUBJECT=Stratum SORT=YES > CORRTYPE=UNSTRUCTURED ADJUSTCORR=YES > COVB=ROBUST MAXITERATIONS=100 PCONVERGE=1e-006(ABSOLUTE) UPDATECORR=1 > /MISSING CLASSMISSING=EXCLUDE > /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION. > > This assumes X is continuous. If it is categorical, change WITH to BY on > the first line. > > This model may not converge either, which would suggest there's some > problem > with your data. But I think it's worth a try. > > HTH. > > > > E. Bernardo wrote >> Thank you Bruce for your comments. >> The warning was "Since coefficients did not converge, no further models >> will be fitted."The Iteration History states that "At least one >> coefficient is tending to infinity after 3 iterations".So increasing the >> number of iteration from 20 to 50 does not help since it stopped at the >> 3rd iteration. Any further suggestion to solve the problem is highly >> solicited. >> >> Thank you. >> >> >> On Wednesday, March 23, 2016 2:04 AM, Bruce Weaver < > >> bruce.weaver@ > >> > wrote: >> >> >> Did you try increasing the number of iterations on the /CRITERIA >> sub-command? >> E.g., >> >> COXREG faketime >> /STATUS=outcome(1) >> /STRATA=Stratum >> /METHOD=ENTER X >> /CRITERIA=PIN(.05) POUT(.10) ITERATE(50). >> >> >> >> >> E. Bernardo wrote >>> Thank you for the link. I think my SPSS syntax now is correct.COXREG >>> faketime /STATUS=outcome(1) /STRATA=Stratum /METHOD=ENTER X >>> /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). >>> However there is warning as part of the outputs. This warning is:"Since >>> coefficients did not converge, no further models will be fitted." >>> And the odds ratio is too large: Exp(B) = 74.769 >>> Is the odds ratio wrong? >>> SPSS outputs are pasted below. >>> Eins >>> >>> >>> | Warnings | | | | | | | | | | >>> | Since coefficients did not converge, no further models will be fitted. >>> | >>> | | | | | | | | | >>> | | | | | | | | | | | >>> | Case Processing Summary | | | | | | | >>> | | N | Percent | | | | | | | >>> | Cases available in analysis | Eventa | 47 | 33.3% | | | | | | | >>> | Censored | 94 | 66.7% | | | | | | | >>> | Total | 141 | 100.0% | | | | | | | >>> | Cases dropped | Cases with missing values | 0 | 0.0% | | | | | | >>> | >>> | Cases with negative time | 0 | 0.0% | | | | | | | >>> | Censored cases before the earliest event in a stratum | 0 | 0.0% | | >>> | >>> | | | | >>> | Total | 0 | 0.0% | | | | | | | >>> | Total | 141 | 100.0% | | | | | | | >>> | a. Dependent Variable: faketime | | | | | | | >>> | | | | | | | | | | | >>> | Stratum Statusa | | | | | | | >>> | Stratum | Event | Censored | Censored Percent | | | | | | | >>> | 1.0 | 1 | 2 | 66.7% | | | | | | | >>> | 2.0 | 1 | 2 | 66.7% | | | | | | | >>> | 3.0 | 1 | 2 | 66.7% | | | | | | | >>> | 4.0 | 1 | 2 | 66.7% | | | | | | | >>> | 5.0 | 1 | 2 | 66.7% | | | | | | | >>> | 6.0 | 1 | 2 | 66.7% | | | | | | | >>> | 7.0 | 1 | 2 | 66.7% | | | | | | | >>> | 8.0 | 1 | 2 | 66.7% | | | | | | | >>> | 9.0 | 1 | 2 | 66.7% | | | | | | | >>> | 10.0 | 1 | 2 | 66.7% | | | | | | | >>> | 11.0 | 1 | 2 | 66.7% | | | | | | | >>> | 12.0 | 1 | 2 | 66.7% | | | | | | | >>> | 13.0 | 1 | 2 | 66.7% | | | | | | | >>> | 14.0 | 1 | 2 | 66.7% | | | | | | | >>> | 15.0 | 1 | 2 | 66.7% | | | | | | | >>> | 16.0 | 1 | 2 | 66.7% | | | | | | | >>> | 17.0 | 1 | 2 | 66.7% | | | | | | | >>> | 18.0 | 1 | 2 | 66.7% | | | | | | | >>> | 19.0 | 1 | 2 | 66.7% | | | | | | | >>> | 20.0 | 1 | 2 | 66.7% | | | | | | | >>> | 21.0 | 1 | 2 | 66.7% | | | | | | | >>> | 22.0 | 1 | 2 | 66.7% | | | | | | | >>> | 23.0 | 1 | 2 | 66.7% | | | | | | | >>> | 24.0 | 1 | 2 | 66.7% | | | | | | | >>> | 25.0 | 1 | 2 | 66.7% | | | | | | | >>> | 26.0 | 1 | 2 | 66.7% | | | | | | | >>> | 27.0 | 1 | 2 | 66.7% | | | | | | | >>> | 28.0 | 1 | 2 | 66.7% | | | | | | | >>> | 29.0 | 1 | 2 | 66.7% | | | | | | | >>> | 30.0 | 1 | 2 | 66.7% | | | | | | | >>> | 31.0 | 1 | 2 | 66.7% | | | | | | | >>> | 32.0 | 1 | 2 | 66.7% | | | | | | | >>> | 33.0 | 1 | 2 | 66.7% | | | | | | | >>> | 34.0 | 1 | 2 | 66.7% | | | | | | | >>> | 35.0 | 1 | 2 | 66.7% | | | | | | | >>> | 36.0 | 1 | 2 | 66.7% | | | | | | | >>> | 37.0 | 1 | 2 | 66.7% | | | | | | | >>> | 38.0 | 1 | 2 | 66.7% | | | | | | | >>> | 39.0 | 1 | 2 | 66.7% | | | | | | | >>> | 40.0 | 1 | 2 | 66.7% | | | | | | | >>> | 41.0 | 1 | 2 | 66.7% | | | | | | | >>> | 42.0 | 1 | 2 | 66.7% | | | | | | | >>> | 43.0 | 1 | 2 | 66.7% | | | | | | | >>> | 44.0 | 1 | 2 | 66.7% | | | | | | | >>> | 45.0 | 1 | 2 | 66.7% | | | | | | | >>> | 46.0 | 1 | 2 | 66.7% | | | | | | | >>> | 47.0 | 1 | 2 | 66.7% | | | | | | | >>> | Total | 47 | 94 | 66.7% | | | | | | | >>> | a. The strata variable is : Stratum | | | | | | | >>> | | | | | | | | | | | >>> | | | | | | | | | | | >>> | Block 0: Beginning Block | | | | | | | | | | >>> | | | | | | | | | | | >>> | Omnibus Tests of Model Coefficients | | | | | | | | | | >>> | -2 Log Likelihood | | | | | | | | | | >>> | 103.270 | | | | | | | | | | >>> | | | | | | | | | | | >>> | | | | | | | | | | | >>> | Block 1: Method = Enter | | | | | | | | | | >>> | | | | | | | | | | | >>> | Iteration Historyb | | | | | | | | >>> | | -2 Log Likelihooda | Coefficient | | | | | | | | >>> | X | | | | | | | | >>> | 1 | 80.334 | 2.071 | | | | | | | | >>> | 2 | 76.835 | 3.255 | | | | | | | | >>> | 3 | 75.745 | 4.314 | | | | | | | | >>> | a. Beginning Block Number 0, initial Log Likelihood function: -2 Log >>> likelihood: 103.270 | | | | | | | | >>> | b. At least one coefficient is tending to infinity after 3 iterations >>> | >>> | | | | | | | >>> | | | | | | | | | | | >>> | Omnibus Tests of Model Coefficientsa | >>> | -2 Log Likelihood | Overall (score) | Change From Previous Step | >>> Change >>> From Previous Block | >>> | Chi-square | df | Sig. | Chi-square | df | Sig. | Chi-square | df | >>> Sig. >>> | >>> | 75.745 | 20.024 | 1 | .000 | 27.524 | 1 | .000 | 27.524 | 1 | .000 | >>> | a. Beginning Block Number 1. Method = Enter | >>> | | | | | | | | | | | >>> | Variables in the Equation | | | | >>> | | B | SE | Wald | df | Sig. | Exp(B) | | | | >>> | X | 4.314 | 1.861 | 5.375 | 1 | .020 | 74.769 | | | | >>> | | | | | | | | | | | >>> | Covariate Means | | | | | | | | | >>> | | Mean | | | | | | | | | >>> | X | .262 | | | | | | | | | >>> >>> >>> >>> On Saturday, March 12, 2016 5:35 PM, Marta Garcia-Granero < >> >>> mgarciagranero@ >> >>> > wrote: >>> >>> >>> Hi >>> >>> I don't think your syntax is correct. You have a create a false " time >>> variable" first , let's call it pseudotime, with values EQ 2 for >>> controls >>> and EQ 1 for cases. This variable acts as time on the COXREG command. >>> Status is defined by CaseControl, the covariable/factor is HealthStatus, >>> and Matchgroup acts as strata variable. >>> >>> Did you try the Mantel-Haenszel statistic approach? >>> >>> Sorry I can't be more detailed/helpful, but I'm at an airport right >>> with just my laptop (away from the desktop office computer with all my >>> old >>> syntax collection. >>> >>> Regards, >>> Marta Garcia-Granero >>> >>> El 12/03/2016 a las 4:36, E. Bernardo escribió: >>> >>> Thank you for your comments, Marta and Rich. >>> Yes, raw data are available. The summary data are just my example to >>> illustrate my data. Actually, I tried to use the C regression (as trick >>> to >>> conduct Conditional regression in SPSS). However, the SPSS reports only >>> the -2LL. I dont know if I did the correct trick. >>> Marta, following the variable names you provided in the previous email, >>> here is my syntax: >>> DATASET ACTIVATE DataSet2. COXREG Casecontrol /STATUS=HeathStatus >>> /STRATA=Matchgroup /CRITERIA=PIN(.05) POUT(.10) ITERATE(20). >>> What is wrong with my syntax? >>> Thank you. >>> >>> On Friday, March 11, 2016 2:27 AM, Rich Ulrich < >> >>> rich-ulrich@ >> >>> > wrote: >>> >>> >>> #yiv5634543243 --.yiv5634543243hmmessage >>> P{margin:0px;padding:0px;}#yiv5634543243 >>> body.yiv5634543243hmmessage{font-size:12pt;font-family:Calibri;}#yiv5634543243 >>> As Marta G-G says, you are very limited if you don't have the raw data. >>> And, abstractly, I'm not grasping how the McNemar Test would fit, even >>> for a >>> simple pair. >>> >>> Given the summary numbers, you have a simple contingency table. >>> If those are the numbers, there is not much doubt about a difference. >>> >>> However, if the "matching" was important, then taking it into account >>> would reduce the size of the apparent effect. That's simple logic. > ("Why >>> do we >>> want to match? - Because those variables might account for the >>> outcome.") >>> >>> Often, trying to match is useful for collecting/selecting Controls to >>> use. However, >>> I have long been hostile to analyzing such data as "matched", given (a) >>> the >>> imprecision of much of such matching ("age within 4 years"); (b) the > loss >>> of >>> d.f. in the analysis; and, (c) the weakness of treating the matching >>> covariates >>> essentially as categories instead of as continuous measures. - The >>> alternative >>> of using the matching variables as covariates (in a logistic >>> here) >>> is almost bound to be more powerful and more robust. - That, in turn, >>> requires >>> that the data be entered with a line for each subject, with the Group, >>> Status, >>> and personal covariates. >>> >>> -- >>> Rich Ulrich >>> >>> >>> >>> Date: Thu, 10 Mar 2016 05:46:45 +0000 >>> From: >> >>> [hidden email] >> >>> Subject: Mc Nemar Test for 1:2 Case-Control ? >>> To: >> >>> [hidden email] >> >>> >>> Dear Members, >>> Consider the 1:2 matched study where each case is matched with 2 >>> controls. Health status (classified as "+" or "-" ) of all cases and >>> controls were recorded. Summary of the gathered data looks like as >>> follows: >>> Group HealthStatus Count Case + 50 >>> Case - 50 Control + >>> >>> 10 Control - 190 >>> Can we use McNemar Test to test if the proportion of cases on the "+" >>> HealthStatus is greater than the controls? Please suggest an appropriate >>> test if McNemar test is not appropriate. >>> Thank you. >>> ===================== To manage your subscription to SPSSX-L, send >>> 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 >> >>> [hidden email] >> >>> (not to SPSSX-L), with no body text except thecommand. To leave the > list, >>> send the commandSIGNOFF SPSSX-LFor a list of commands to manage >>> subscriptions, send the commandINFO 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 >> >> >> >> >> >> ----- >> -- >> Bruce Weaver > >> bweaver@ > >> 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. >> >> -- >> View this message in context: >> http://spssx-discussion.1045642.n5.nabble.com/Mc-Nemar-Test-for-1-2-Case-Control-tp5731696p5731793.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 > > > > > > ----- > -- > Bruce Weaver > bweaver@ > 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. > > -- > View this message in context: > http://spssx-discussion.1045642.n5.nabble.com/Mc-Nemar-Test-for-1-2-Case-Control-tp5731696p5731801.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 ----- -- 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. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Mc-Nemar-Test-for-1-2-Case-Control-tp5731696p5731838.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 |
Re: Mc Nemar Test for 1:2 Case-Control ?
|
Administrator
|
Thanks. Given your results (OR = 1.000067, p < .0001), it would appear that your sample size is very large, because you are getting a very low p-value for an OR that is nowhere near clinical significance. What is the N?
--
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: Mc Nemar Test for 1:2 Case-Control ?
|
In reply to this post by E. Bernardo-2
Such a coarse DV!
Many populations would always have some pain.
Art Kendall
Social Research Consultants |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |