No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression
Locked
 
|
So we are having problems with our regression.
Ordinal Logistic Regression: - The dependent, in our case the Health Literacy Score (ordinal date (low, intermediate, high) ( all data is there, no missing values) - Risk factors still gives no valid case were found when doing this regression, with the covariates(ex. age) and factors correct ( ex. motivation). We sometimes do have missing values in these covariates or factors, or for some of them a lot of people have the same answer in this. Is that why we keep having the same message ' no valid cases were found?' Multivariable lineair regression - the dependent, in our case again the health literacy score, but now in a continuous way. - risk factors were okay (we deleted the ones that usually had the same answer in 90% of the cases for example, the person was married in 90% of the case) BUT => we did have B-values , but no P-values (sig) in the tables of the regression. Can anybody help me with this? We keep getting stuck and its for our master thesis about health literacy. |
Re: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression
|
|
Multiple analyses with maybe same/maybe different problems.
Ordinal regression. Have you done a careful, thorough missing data analysis? Maybe you have but you don't say that you have. I suggest you do these things. 1) Count the number of cases with missing values for ALL variables in the problem analysis. Do you have cases with all analysis variables having valid data? 2) Crosstab that number against the dependent variable (DV). Does every value of the DV have valid cases? What's the smallest number of valid cases. 3) If you select those cases with no missing data, are you absolutely confident that EVERY independent variable has some cases in every category of your DV? Meaning have you done the analysis to c heck? Please explain this statement."... for some of them a lot of people have the same answer in this." Linear regression. What does this statement mean "... we deleted the ones that usually had the same answer in 90% of the cases for example, the person was married in 90% of the case)." The sentence construction is really confusing but it seems as if you deleted cases. Let's say you did delete cases. I think you should begin with a missing data analysis to make sure you cases with valid data. I don't ever recall seeing a regression output with B values but no p values. So I don't know what specifically is wrong. Let's do diagnostics. You have B values. Do you have SEs? Do you have beta values? Do you have t values? Do you have anything in the p value column? Any numbers at all? In your reply would you please copy the coefficients table. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of annekecl Sent: Sunday, February 21, 2016 11:46 AM To: [hidden email] Subject: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression So we are having problems with our regression. Ordinal Logistic Regression: - The dependent, in our case the Health Literacy Score (ordinal date (low, intermediate, high) ( all data is there, no missing values) - Risk factors still gives no valid case were found when doing this regression, with the covariates(ex. age) and factors correct ( ex. motivation). We sometimes do have missing values in these covariates or factors, or for some of them a lot of people have the same answer in this. Is that why we keep having the same message ' no valid cases were found?' Multivariable lineair regression - the dependent, in our case again the health literacy score, but now in a continuous way. - risk factors were okay (we deleted the ones that usually had the same answer in 90% of the cases for example, the person was married in 90% of the case) BUT => we did have B-values , but no P-values (sig) in the tables of the regression. Can anybody help me with this? We keep getting stuck and its for our master thesis about health literacy. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/No-Valid-Case-Were-found-Ordinal-Logistic-Regression-and-multivariable-lineair-regression-tp5731551.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 |
|
|
In reply to this post by annekecl
I presume that you are using the PLUM procedure. PLUM uses listwise deletion of cases with missing values. You can specify /MISSING=INCLUDE (not supported in the dialog box) to include user missing values as valid, but system missing value are never included. So most likely all of the cases have a missing value on at least one variable. You could use Missing Value Analysis (on the Analyze menu) if you have this option to investigate. Alternatively you could use the NMISS function applied to your independent variables to get a count of the missing values in each case. On Sun, Feb 21, 2016 at 9:46 AM, annekecl <[hidden email]> wrote: So we are having problems with our regression. |
Re: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression
|
|
In reply to this post by annekecl
I think that the "B-values with no P-values" indicates that you do not have
=====================
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
enough cases to have a full-rank covariance matrix. There are ploys to use to get B-values in that sort of case, but the error is infinite. Here is a question: Why does one analysis say NO CASES and then other does find a few? Are you specifying the PLUM wrong? What do you have for (1) total N; (2) total number of variables; (3) range of proportion-missing among the predictors? If you want robust tests and coefficients in Ordinary Least Squares regression, you should hope to have (say) 10 times the number of valid cases as you have predictors. The prescription is a higher ratio for Logistic -- like, 20 times the smaller category. I have dropped a variable as being "probably non-informative" when it had as few as 3 cases varying ... but that was a far smaller fraction than "only 10%". On the other hand, I suspect that you need to drop /most/ of your predictors, given your sample size. Or, would your sample size for the analysis become big, if you replaced all Missing with some estimated value? Perhaps you should consider how to construct "composite scores" that you combine similar sets of variables, and overcome the Missing Values problem by creating Composites that seem reasonable and valid, even when created with one or more missing items: Such as, taking an average of several scaled items that are present; or the maximum of several. -- Rich Ulrich > Date: Sun, 21 Feb 2016 09:46:14 -0700 > From: [hidden email] > Subject: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression > To: [hidden email] > > So we are having problems with our regression. > > Ordinal Logistic Regression: > - The dependent, in our case the Health Literacy Score (ordinal date (low, > intermediate, high) ( all data is there, no missing values) > - Risk factors still gives no valid case were found when doing this > regression, with the covariates(ex. age) and factors correct ( ex. > motivation). > We sometimes do have missing values in these covariates or factors, or for > some of them a lot of people have the same answer in this. > Is that why we keep having the same message ' no valid cases were found?' > > Multivariable lineair regression > - the dependent, in our case again the health literacy score, but now in a > continuous way. > - risk factors were okay (we deleted the ones that usually had the same > answer in 90% of the cases for example, the person was married in 90% of the > case) > > BUT => we did have B-values , but no P-values (sig) in the tables of the > regression. > > Can anybody help me with this? > We keep getting stuck and its for our master thesis about health literacy. > > |
|
|
In reply to this post by Jon Peck
This seems like a pretty valid theory.. Because we do have missing values for a lot of the covariates or factors (risk factors in our case). But were do I put in the /Missing=include ? Could you show me a screenshot of that? Cause I think this might solve the whole problem...
NMISS Function? Date: Mon, 22 Feb 2016 08:37:53 -0700 From: [hidden email] To: [hidden email] Subject: Re: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression I presume that you are using the PLUM procedure. PLUM uses listwise deletion of cases with missing values. You can specify /MISSING=INCLUDE (not supported in the dialog box) to include user missing values as valid, but system missing value are never included. So most likely all of the cases have a missing value on at least one variable. You could use Missing Value Analysis (on the Analyze menu) if you have this option to investigate. Alternatively you could use the NMISS function applied to your independent variables to get a count of the missing values in each case. On Sun, Feb 21, 2016 at 9:46 AM, annekecl <[hidden email]> wrote: So we are having problems with our regression. If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/No-Valid-Case-Were-found-Ordinal-Logistic-Regression-and-multivariable-lineair-regression-tp5731551p5731568.html
To unsubscribe from No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression, click here. NAML |
RE: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression
Administrator
|
Please consult the FM for the placement of /MISSING=INCLUDE!!
It is a SUBCOMMAND in MANY procedures. COMPUTE Cases_with_Missing=NMISS(variable list). FREQUENCIES Cases_with_Missing. It is not a terrible idea to acquaint yourself with your data and the analytical methods you are utilizing prior to conducting analyses! GIGO!!!
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me. --- "Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis." Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?" |
|
|
In reply to this post by annekecl
David has pointed out how to specify the missing values behavior - just paste the syntax if you are using the dialog box interface and then insert it before the command ending period. However, you need to consider whether this treatment makes sense for your data. For scale variables, it almost surely does not, although it might be okay for factors. On Mon, Feb 22, 2016 at 11:44 AM, annekecl <[hidden email]> wrote:
|
|
|
In reply to this post by David Marso
I am not known with SPSS at all, we used a different program in my statistic course..
So what do you mean with the FM? I get it's a subcommand and you can fill it in somewhere, but I don't seem to figure out where .. Is it in my output screen? I am kind of getting lost .. |
RE: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression
Administrator
|
FM = the final two letters of RTFM, which you can Google. But note that in this forum, the F in FM is usually said to stand for FINE. ;-)
To open a PDF of the FM, click on Help > Command Syntax Reference. 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: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression
|
|
In reply to this post by annekecl
What program do you use in your stats class? Just curious, now, are you using spss for your project? If you are not using spss for your project and you had problems with the stats program you are using, why post on this spss listserv AND not disclose that you are using another program. Gene Maguin
-----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of annekecl Sent: Monday, February 22, 2016 2:33 PM To: [hidden email] Subject: Re: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression I am not known with SPSS at all, we used a different program in my statistic course.. So what do you mean with the FM? I get it's a subcommand and you can fill it in somewhere, but I don't seem to figure out where .. Is it in my output screen? I am kind of getting lost .. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/No-Valid-Case-Were-found-Ordinal-Logistic-Regression-and-multivariable-lineair-regression-tp5731551p5731574.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 |
|
|
We used statistica, but have to use SPSS now because our thesis promotors know this program better. We are also doing analysis as the ordinal regression and the multiple lineair regression that we have never even learned about, hence the problems we are experiencing. We all study pharmacy, so aren't statistic geniuses at all.. We know the base of stats. Verstuurd vanaf mijn iPhone Op 22 feb. 2016 om 22:22 heeft Maguin, Eugene [via SPSSX Discussion] <[hidden email]> het volgende geschreven:
|
Re: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression
|
Administrator
|
Well you know what they (used to) say about SPSS: Real Stats. Real Easy. ;-)
http://spssx-discussion.1045642.n5.nabble.com/Real-Stats-Real-Easy-td5722880.html
--
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: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression
|
|
In reply to this post by Rich Ulrich
[Your post to me was not directed to the SPSS Group. I am sharing it with
this Reply, assuming that this was an accident since you would have received two versions from me, one from the List, and the other as email.] I will comment in reverse order, which starts with the simpler answers. For a categorical variable with multiple categories and some Missing, you can move to a dichotomy. For "married", it is common to use a dichotomy, "Marked as currently married" vs "Other"; or "Marked as ever married" vs "Other". You remove "Missing" by definition. Range of proportion missing: X % have no missing values (What is X?); for the others, you might be able to say that most(?) are missing only one or two items, ranging up to X% missing for <something-that-you-still want to look at>. PLUM: I haven't been using it; if it expects categories to be coded 1-3, then it would leave out all cases with "0" or "4" if those were among your codes. With 35 variables, you have a robust regression with 3 or 4 predictors for the ordinary regression, and 2 or 3 for the logistic. If your data are really nice and uncorrelated (relatively speaking), you might be able to have decent coefficients (and tests) for the OLS regression with as many as 7 or 8 variables; fewer, for PLUM. Faced with data like that, I think that I would take my 3 or 4 predictors that seem essential for a model, and create many models that always use them, and use a subset of what is left. However, instead of considering 40 predictors and their 40 hypotheses as separate, I would make use of composite scores (as mentioned in my previous note) after doing a coarse, arbitrary, subjective sort that divides all variables into More vs Less important on a-priori grounds: Your theory and what you have read in the literature. -- Rich Ulrich From: [hidden email] To: [hidden email] Subject: RE: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression Date: Mon, 22 Feb 2016 19:40:57 +0100 If the error is infinite, then that's not a good indication of the conclusion of those risk factors I suppose? The thing is, I have severel factors and covariates (about 30, maybe even more in total). and what do you mean with specifying the PLUM wrong? total patients is 35 in this group, the number of covariates and factors are 42. Range of proportion- missing among the predictors? what do you mean with that? the amount of missing values in the 42 covariates? Because we filled in no value if they didn't answer the question. But is that wrong? And we cannot work with estimated values for somethings, for example we can't predict if the patient in question is married, widower, single , ... From: [hidden email] To: [hidden email]; [hidden email] Subject: RE: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression Date: Mon, 22 Feb 2016 12:50:33 -0500 I think that the "B-values with no P-values" indicates that you do not have enough cases to have a full-rank covariance matrix. There are ploys to use to get B-values in that sort of case, but the error is infinite. Here is a question: Why does one analysis say NO CASES and then other does find a few? Are you specifying the PLUM wrong? What do you have for (1) total N; (2) total number of variables; (3) range of proportion-missing among the predictors? If you want robust tests and coefficients in Ordinary Least Squares regression, you should hope to have (say) 10 times the number of valid cases as you have predictors. The prescription is a higher ratio for Logistic -- like, 20 times the smaller category. I have dropped a variable as being "probably non-informative" when it had as few as 3 cases varying ... but that was a far smaller fraction than "only 10%". On the other hand, I suspect that you need to drop /most/ of your predictors, given your sample size. Or, would your sample size for the analysis become big, if you replaced all Missing with some estimated value? Perhaps you should consider how to construct "composite scores" that you combine similar sets of variables, and overcome the Missing Values problem by creating Composites that seem reasonable and valid, even when created with one or more missing items: Such as, taking an average of several scaled items that are present; or the maximum of several. -- Rich Ulrich > Date: Sun, 21 Feb 2016 09:46:14 -0700 > From: [hidden email] > Subject: No Valid Case Were found - Ordinal Logistic Regression and multivariable lineair regression > To: [hidden email] > > So we are having problems with our regression. > > Ordinal Logistic Regression: > - The dependent, in our case the Health Literacy Score (ordinal date (low, > intermediate, high) ( all data is there, no missing values) > - Risk factors still gives no valid case were found when doing this > regression, with the covariates(ex. age) and factors correct ( ex. > motivation). > We sometimes do have missing values in these covariates or factors, or for > some of them a lot of people have the same answer in this. > Is that why we keep having the same message ' no valid cases were found?' > > Multivariable lineair regression > - the dependent, in our case again the health literacy score, but now in a > continuous way. > - risk factors were okay (we deleted the ones that usually had the same > answer in 90% of the cases for example, the person was married in 90% of the > case) > > BUT => we did have B-values , but no P-values (sig) in the tables of the > regression. > > Can anybody help me with this? > We keep getting stuck and its for our master thesis about health literacy. > > |
|
|
There is another approach that might work in this case and could deal with both an ordinal dependent and a surplus of variables. Analyze > Regression > Optimal Scaling, which is the CATREG procedure, can construct an ordinal or spline ordinal measure for the dependent variable - probably ordinal would be appropriate in this case. It can impute missing values either using the mode or as an extra category. Then, using the lasso regularization, it can in effect select the most important variables, since lasso tends to shrink unimportant variables to a zero coefficient. You can use crossvalidation to select the optimal shrinkage parameter. This is part of the Categories option. If you use it, be careful with the variable coding. Your categorical variables must have positive integer values. On Mon, Feb 22, 2016 at 5:52 PM, Rich Ulrich <[hidden email]> wrote:
|
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |