What is the best "Working correlation matrix" for a study in eye numbers, such as 75 eyes (50 persons) using Chi-square, that were not measured over time? Thank you for the best-detailed clip of GEE using SPSS.
Such a great lecture! Thanks for sharing. I have a question regarding controlling variables. If there were multiple repeat measurement variables(e.g. weight, height), should I also do interaction between the visits and these variables in the same model for adjustment, or just put these variables into the model? Thank you.
I think you should run Pearson correlation analysis and select a structure which is close to a observed correlation. To compute Pearson corr. you need to transform data from longitudinal to a horizontal format, though.
Hello Ayumi, thank you for this clear video. I have a question regarding controlling for variables. How do you do this in GEE? I have 10 variables and 5 of them I want to control for to check what the remaining 5 are contributing to my independent variable. Woudl love to hear from you.
At 6:26 of this video, you see Covariates and Factors boxes in SPSS. You can put numeric variables into Covariates box and categorical variables into Factor box to control for in GEE.
You can include any covariates as you do in other models in SPSS. Slide 6:36 for example, put categorical variable in Factor box, continuous variable in Covariate box.
Thank you for your very helpful video. I wanna use GEE for the analysis but I face this message 'There are at least two records with the same values for the subject and within-subject variables. No output will be displayed.' Could you help me with what I should do?
I found this in SPSS web-page, I hope this helps. www.ibm.com/support/pages/warning-levels-repeated-effect-are-not-different-each-observation-within-repeated-subject
Excellent. On the 21.21 you mean that we're looking for larger p-value, not smaller as usual, right? Small p-value would mean that the interaction compare to the reference interaction is the same, that actually means there is no significant difference in interactions, right?
P-value for the interaction for example for time 2 x treatment assess if OR of treatment at time 2 is different from OR of treatment at the reference time. In the example, as P>0.05 there is no evidence to indicate an interaction.
Hi Mohamed. The data in the example were in long format. If your data are in wide format, you can use Restructure in SPSS under the Data drop down menu. Here is a useful site for tutorials in statistics showing how to restructure wide to long using SPSS syntax, stats.idre.ucla.edu/spss/modules/reshaping-data-wide-to-long-in-versions-11-and-up/ All the best.
Hi! Nice video. Thanks for sharing! Quick question. My analyze drop menu does not have the Generalized Linear Models option. I think it is because I have a SPSS Statistic Base Version. If that is the case, which version do I need in order to be able to run this analysis? Thank you,
Hello. The topic is very interesting, although the flow a bit chaotic, it lacks an order. It can improve but I left a like for taking the time to explain all that :-)
Sara Alotaibi 1 day ago Hi, I have a question on how to run a GEE with Ordinal Logistic model and be able to include the 'EM means' in the output. My study data-set includes ordinal data (a 5-point scale ranging from 1 highly unfamiliar to 5 highly familiar).I’ve run a GEE in SPSS and chose an ‘ordinal logistic’ model but it’s not possible to include ‘EM means’ in this type of model and for this type of data. Thus, when I get the results, no ‘pairwise comparisons’ table is included!
Hello. I try to find "nonparemetric repeated-measures ANOVA in SPSS ". Could I do it with "Generalized Estimating Equation (GEE) in SPSS " ? There are two groups and pretest, posttest, retention tests were applied.
You should not use ANOVA as it requires independence in data. You do not want to use repeated measured ANOVA as it uses a horizontal data format which deletes the entire set of observations for a single patients with at least one outcome measure missing. You can of course use mixed model for repeatedly measured continuous outcome. GEE is more robust than mixed model if you specify a wrong variance-covariance structure. But Mixed model is better if data contain missing outcomes. Please see my video for missing data imputation.
7:49 Something about the slope explanation doesn't sit right here..... Its main effect of age i.e., the slope a scatter plot with age and dental growth with pooled data [Boys and girls together]. How is it that 0.49 is the slope of only the girls and we add 0.294 from interaction to get the slope of boys? Main effect doesn't distinguish the gender category. Can you confirm?
When the interaction term is included in the model along with the main effect, A+B+A*B the regression coefficient for the main effect A only estimates the effect of the variable A among among the reference category of the other variable B (sex in this case, girls is ref in this case).
Thank you for the kind vedio and explanations. I have a similar question as Ganne Chaitanya, too. 7:24 1. What does the main effect of sex (B=-1.044) mean in this setting? 2. In your video and answer, my understanding of the regression coefficient A+B+A*B means A= age, B=sex, A*B= age*sex, why was regression coefficient B omitted in this situation? Very much expect and appreciate your answer.
With binary data, OR gives you an idea about the effect size. With continuous data, you can use Cohen's f^2 = r^2 / (1 - r^2), interpreted as .02=small, .15=medium, .35=large. See Cohen (2003) "A Power Primer" for details.
When visit is included in the model as a factor variable, it means that Treatment effect if greater at visit4 relative to that at the reference visit. Let's say if OR for treatment at visit 4 is 3.0 and that for the reference visit i in the model as a factor variable, it means that Treatment effect if
@@ayumishintani7044 alright understood. I have 1 more question. If i have various variable such as those in sociodemographics characteristics, should i include all that variable in the GEE model so that the model i obtained in the GEE analysis somehow being adjusted to that variables (as covariates) as well? if i can do that, should i just include all other variable or include the one with significant at bivariate analysis?
@@Capiel89 Bivariate selection is not recommended. You should decide which variable to include without looking at the data among risk factors for the outcome variable.
@@ayumishintani7044 alright understood . However i need to ask about how to properly report the finding based on the interpretation for the significant interaction for example Age*time. Since it is significant, should i write, we can conclude that "The dental growth for boys is 0.294 higher compared to girls, over time"?
@@Capiel89 I assume you are referring to the result of the analysis I explained in 13:44 "Lesson 29 (5) Mixed Effect Model Comparing 2 slopes", yes your explanation is correct.
Wonderful! Really helpful introduction into longitudinal GEE. Thanks for putting this to TH-cam
Prof Ayumi Shintani, thank you for this informative video on GEE.
Thank you for the precise and direct explanation! It was exactly what I needed.
Your VDO is really good and can helped me a lot. Keep do it. I will follow your work. I am very appreciate that. Thank you so much.
Great presentation! Very Informative. Thanks for sharing!
Your explanations really helped me with my paper! Keep up the good work!
Super, very helpful. Thanks for your time preparing this.
Very clear :) Please keep them coming, Very valuable videos.
Thank you very much Sensei, very helpful and simple explanation.
Glad it was helpful!
Thank you for this lecture professor! Would you be able to post the resp.sav for us to practice ?
Thank you. How can i control confounding variables with GEE?
This is such a helpful video, thank you!
HI
What is the best "Working correlation matrix" for a study in eye numbers, such as 75 eyes (50 persons) using Chi-square, that were not measured over time? Thank you for the best-detailed clip of GEE using SPSS.
I want to know how to draw a graph and how to report the findings in a dissertation. Thanks for your information.
In your example, which is the cluster variable? Thank you for the video!
Cluster variable is patient ID.
Such a great lecture! Thanks for sharing. I have a question regarding controlling variables. If there were multiple repeat measurement variables(e.g. weight, height), should I also do interaction between the visits and these variables in the same model for adjustment, or just put these variables into the model? Thank you.
Thank you so much for this valuable video. Please which Working Correlation Matrix is suitable when there is missing data?
Regards,
GEE simply drops the observations with missing data. You can use the QIC criterion in correlation structure selection.
I think you should run Pearson correlation analysis and select a structure which is close to a observed correlation. To compute Pearson corr. you need to transform data from longitudinal to a horizontal format, though.
thank you for the video. could you please answer how to use designed effect while using GEE analysis for cluster randomized trial
You can use exchangeable structure to deal with data clusters.
Hello Ayumi, thank you for this clear video. I have a question regarding controlling for variables. How do you do this in GEE? I have 10 variables and 5 of them I want to control for to check what the remaining 5 are contributing to my independent variable. Woudl love to hear from you.
At 6:26 of this video, you see Covariates and Factors boxes in SPSS. You can put numeric variables into Covariates box and categorical variables into Factor box to control for in GEE.
Thank you for the video. Very useful, how do you adjust for different variables, let us we want to adjust for sex and age?
You can include any covariates as you do in other models in SPSS. Slide 6:36 for example, put categorical variable in Factor box, continuous variable in Covariate box.
It is very valuable thank you
Thank you for your very helpful video. I wanna use GEE for the analysis but I face this message 'There are at least two records with the same values for the subject and within-subject variables. No output will be displayed.' Could you help me with what I should do?
I found this in SPSS web-page, I hope this helps. www.ibm.com/support/pages/warning-levels-repeated-effect-are-not-different-each-observation-within-repeated-subject
Excellent. On the 21.21 you mean that we're looking for larger p-value, not smaller as usual, right? Small p-value would mean that the interaction compare to the reference interaction is the same, that actually means there is no significant difference in interactions, right?
P-value for the interaction for example for time 2 x treatment assess if OR of treatment at time 2 is different from OR of treatment at the reference time. In the example, as P>0.05 there is no evidence to indicate an interaction.
Thanks! how do you structure the data before doing these steps? do you mind sharing a sample?
Hi Mohamed. The data in the example were in long format. If your data are in wide format, you can use Restructure in SPSS under the Data drop down menu. Here is a useful site for tutorials in statistics showing how to restructure wide to long using SPSS syntax, stats.idre.ucla.edu/spss/modules/reshaping-data-wide-to-long-in-versions-11-and-up/ All the best.
@@stuartwatson4102 Thanks Stuart! that's very helpful!
The background noise is very annoying and makes it hard to focus on the content... Too bad, because it's very interesting.
Hi! Nice video. Thanks for sharing!
Quick question. My analyze drop menu does not have the Generalized Linear Models option. I think it is because I have a SPSS Statistic Base Version. If that is the case, which version do I need in order to be able to run this analysis?
Thank you,
Hello. The topic is very interesting, although the flow a bit chaotic, it lacks an order. It can improve but I left a like for taking the time to explain all that :-)
Thanks!
please how can I get the eta square from GEE analysis ?
Can you provide any guidance on the interpretation of estimated marginal means in a GEE with binary outcomes
Is it correct that dummy coding categorical variables is not necessary for GEE because SPSS does it automatically?
yes, by putting any variables in Factor box in SPSS it creates dummy variables automatically.
could it be used for ordinal dependent variables in a repeated measures model ?
Yes, you can.
Hi, how do i decide between robust estimator and model based estimator (under covariance matrix)?
I usually use a robust estimator.
@@ayumishintani7044 thank you! great video
Sara Alotaibi
1 day ago
Hi,
I have a question on how to run a GEE with Ordinal Logistic model and be able to include the 'EM means' in the output.
My study data-set includes ordinal data (a 5-point scale ranging from 1 highly unfamiliar to 5 highly familiar).I’ve run a GEE in SPSS and chose an ‘ordinal logistic’ model but it’s not possible to include ‘EM means’ in this type of model and for this type of data. Thus, when I get the results, no ‘pairwise comparisons’ table is included!
That could be true as Ordinal logistic model does not use means, it uses ranks of data.
I want to this dataset, tell me please.
Hello. I try to find "nonparemetric repeated-measures ANOVA in SPSS
". Could I do it with "Generalized Estimating Equation (GEE) in SPSS
" ?
There are two groups and pretest, posttest, retention tests were applied.
Try running gee for an ordinal outcomes.
Why not put of covariate age ?
Why did you choose this analysis instead of Anova or linear mixed model. When this analysis is more preferable than others?
You should not use ANOVA as it requires independence in data. You do not want to use repeated measured ANOVA as it uses a horizontal data format which deletes the entire set of observations for a single patients with at least one outcome measure missing. You can of course use mixed model for repeatedly measured continuous outcome. GEE is more robust than mixed model if you specify a wrong variance-covariance structure. But Mixed model is better if data contain missing outcomes. Please see my video for missing data imputation.
@@ayumishintani7044 thank u for your answer.
thanks
7:49 Something about the slope explanation doesn't sit right here..... Its main effect of age i.e., the slope a scatter plot with age and dental growth with pooled data [Boys and girls together]. How is it that 0.49 is the slope of only the girls and we add 0.294 from interaction to get the slope of boys? Main effect doesn't distinguish the gender category. Can you confirm?
When the interaction term is included in the model along with the main effect, A+B+A*B the regression coefficient for the main effect A only estimates the effect of the variable A among among the reference category of the other variable B (sex in this case, girls is ref in this case).
Thank you for the kind vedio and explanations. I have a similar question as Ganne Chaitanya, too.
7:24
1. What does the main effect of sex (B=-1.044) mean in this setting?
2. In your video and answer, my understanding of the regression coefficient A+B+A*B means
A= age, B=sex, A*B= age*sex, why was regression coefficient B omitted in this situation?
Very much expect and appreciate your answer.
How can I calculate effect size in this case?
????
With binary data, OR gives you an idea about the effect size. With continuous data, you can use Cohen's f^2 = r^2 / (1 - r^2), interpreted as .02=small, .15=medium, .35=large. See Cohen (2003) "A Power Primer" for details.
after we include the interaction between the treatment and visit, if say the treatment1*visit4 has p
When visit is included in the model as a factor variable, it means that Treatment effect if greater at visit4 relative to that at the reference visit. Let's say if OR for treatment at visit 4 is 3.0 and that for the reference visit i in the model as a factor variable, it means that Treatment effect if
@@ayumishintani7044 alright understood. I have 1 more question. If i have various variable such as those in sociodemographics characteristics, should i include all that variable in the GEE model so that the model i obtained in the GEE analysis somehow being adjusted to that variables (as covariates) as well? if i can do that, should i just include all other variable or include the one with significant at bivariate analysis?
@@Capiel89 Bivariate selection is not recommended. You should decide which variable to include without looking at the data among risk factors for the outcome variable.
@@ayumishintani7044 alright understood . However i need to ask about how to properly report the finding based on the interpretation for the significant interaction for example Age*time. Since it is significant, should i write, we can conclude that "The dental growth for boys is 0.294 higher compared to girls, over time"?
@@Capiel89 I assume you are referring to the result of the analysis I explained in 13:44 "Lesson 29 (5) Mixed Effect Model Comparing 2 slopes", yes your explanation is correct.
What the hell are doing while talking?
Very nice video, but the audio sounds like you're hugging a balloon.
Respected I sent you an email please reply
If you do not mind, would you please post your question here, I cannot find your e-mail.