Hello everyone! It has come to my attention that a few of the links to the datasets, etc. under the video description were broken. I have updated these (as of March 18, 2024). My apologies (and honestly, I don't know how this happened). Additionally, I wanted to let you know that I also have a more recent video on this topic that also demonstrates the use of cluster-robust standard errors. Please go to czcams.com/video/NI2SuLGissc/video.html to view . As always, thank you for visiting!
Thank you for sharing your knowledge with the world. I had trouble with panel data analysis in spss until I found your lecture. The example used to demonstrate the application of fixed effects regression on panel data is very useful. Keep posting more similar videos.
Thank you for this video!!! Surely it's one of the best videos you continually devote your time to create and share with us! It's really makes difference.
Thank you for this video with step-by-step instruction. It was really helpful. Initially, I did the same via the mixed models` option but did not get the model fit. Now I have everything that I need.
Thank you so much for this video! I was wondering: you do create a dummy variable for each characteristic. When running the regression, would I need to use one less dummy variable than I have? To avoid multicollinearity? That's what I understood from reviewing info on OLS models. Would so much appreciate your answer!
Thanks a lot for the video. Maybe I understood something wrong. Why can you just use the tools for a normal multiple regression for panel data without controlling for the time or that the different measurements over time for each state do probably have something in comen?
Thanks for this great video. Is it possible to use a stepwise approach to adding variables in blocks/steps where the dummy-coded variables would enter on their own? If so, would they enter first, last, or in some other temporal fashion in alignment with our theoretical framework/literature that guides the research?
In my case, I'm using undergraduate student-level data, and I'm trying to account for differences across institutions. I have 16 institutions, and I'm selecting one as a reference group to leave out of the regression.
Hi Kaitlyn, If I am reading your question correctly, you are asking if you an enter the dummy variables first in an initial step in the first block, and then in the second block of variables set this to stepwise (to let the algorithm select the best set of within-case predictors). I have never done this myself or seen anyone else use this particular strategy before (truthfully, I am not a big fan of using stepwise approaches AND I have not really spent much time reflecting on this particular question before). As far as I can tell, there is nothing that would seemingly prohibit you from being able to do what you are describing in SPSS (I was able to accomplish it using the dataset from this video). In other words, Step 1 I entered the dummy variable with the entry method to ENTER; in Step 2, I entered the other predictors and reset the entry method to STEPWISE. Please be clear though, I am not endorsing or recommending this approach, but just telling you how it might be accomplished using SPSS. I hope this helps! Cheers!
Thanks for this video and your other great videos, Mike. Are you able to give any pointer on how to control for between-group variation using fixed effects dummy variables when the number of categories of the would-be dummy variable is very high? For example, country-level fixed effects when the number of countries is around 200. How might you create such a dummy variable in SPSS? Thanks again.
Suppose I have 500 cases measured over and over. Is it possible to create dummies for all these and use them in the model? I'm wondering how this model will look like!! Thank you
Hi there. And thanks for the question. When you declare data to be panel data (such as in Stata if you are running a fixed-effects regression), the program takes care of various pre-processing steps prior to running the analysis. In theory, there are two different pre-processing steps can be used with our data when running a fixed-effects regression: Mean-deviation centering method or the creation of dummy coded variables (both methods assume your data are in long/vertical format). The default in Stata (I'll get to SPSS soon) when you use xtset along with the xtreg command and fe option is to recognize that panel data is being analyzed. As a result, Stata takes care of the pre-processing for you (using mean-deviation approach; see Allison, 2009) - and adjusts standard errors in your regression accordingly. If you are using the reg command in Stata, then the program would not recognize the data as panel. In this case, you would have to create dummy variables and then enter them into your regression (or, if you use the i. prefix, this can be used with your factor variable (which will lead to the automatic inclusion of the dummy variables in your regression). The latter method (Least-squares dummy variable approach) is essentially what we are doing in the example in my presentation with SPSS. BOTH SPSS and Stata have the capability to use GEE (Generalized Estimating Equations) with longitudinal data (in fact, I have a video on it in Stata here: czcams.com/video/4vQfuIRIahY/video.html). In Stata, you would be using the xtreg command with pa (for population-averaged), whereas in SPSS you would be going through Generalized linear models --> Generalized estimating equations. I am unclear as to whether you are wondering about SAS or R; but the logic I use above is the same. I can't speak much to SAS, but I do believe the 'plm' package for R uses the same approach (mean-deviations approach) as I noted above for SPSS. I hope this answers your question!
The R-square overall in Stata is computed differently than the R-square in spss, and I don't find the Stata overall R-square particularly useful. But I just created a new presentation on panel regression in spss with repeated measures data you might find helpful at czcams.com/video/cN_4RoeKlT4/video.html
Hello ! Very interesting video! I am a phd student and would like to know if I can contact you personally via email to ask a question concerning Fixed Effects models. Thank you
so nice and straight to the point. how does one chose between fixed effect and random effect using spss. how do does one know which model is best fit between fixed effect and random effect using spss in order to conduct a panel data analysis?
Hi Haiying, it actually isn't different at all mathematically. It's basically a multiple regression using dummy coding of a group membership variable to control for any of the association between the iv's and dv that may be occurring between groups. By controlling for the variation occurring between groups, you end up with the within-groups associations remaining. Here is a link to a presentation that you might find helpful (of course the discussion is mainly in terms of Stata, but it does provides useful conceptual information): www.princeton.edu/~otorres/Panel101R.pdf
@@mikecrowson2462 Thank you very much professor! Does this mean that if you want to estimate time-fixed effects, you can also just simply add the time variable to the multiple linear regression (e.g., year) which would be just the same as using a time-fixed effects model?
The fixed effects model does not assume random intercepts or slopes as you would see in hierarchical models. And it does not allow you to model between-group predictors of the outcome as you could do in the case of a random effects model. SPSS does not have a random effects mechanism as far as I've found. Assuming a reasonably large number of groups randomly sampled from the population of groups (level 2 units in multilevel terminology), I generally prefer to use a multilevel modeling approach. But this fixed effects approach is pretty good when you have a smaller number of level 2 units. If you have predictors you wish to model at the group level (level 2), it is possible to include them only by modeling interactions involving level 1 predictors. Here's a nice article I found that you might find useful: biol609.github.io/Readings/McNeish_Kelley_PsychMethods_2019.pdf
Sir Mike, we are Grateful. Continue making a Difference in the world !
Hello everyone! It has come to my attention that a few of the links to the datasets, etc. under the video description were broken. I have updated these (as of March 18, 2024). My apologies (and honestly, I don't know how this happened). Additionally, I wanted to let you know that I also have a more recent video on this topic that also demonstrates the use of cluster-robust standard errors. Please go to czcams.com/video/NI2SuLGissc/video.html to view . As always, thank you for visiting!
Very Very useful. Brilliant explanation
Thank you for sharing your knowledge with the world. I had trouble with panel data analysis in spss until I found your lecture. The example used to demonstrate the application of fixed effects regression on panel data is very useful. Keep posting more similar videos.
Thanks, I thought SPSS has no capacity for panel data analysis but with this vedio am informed.
Thank you for this video!!! Surely it's one of the best videos you continually devote your time to create and share with us! It's really makes difference.
Thanks for explaining the fixed effect model. Very clear and helpful!
Thank you for this video with step-by-step instruction. It was really helpful. Initially, I did the same via the mixed models` option but did not get the model fit. Now I have everything that I need.
you are such a good teacher.......bravo....very clear and vital information.....will be glad if i can get to connect with you
Excellent. Exactly what I needed. Thanks.
Outstanding video and explanation. Thank you, Mike!
Mike, thanks for explaining this process in SPSS!
Greetings from NL
Thanks for this amazing video. appreciated.
Thank you.
Thanks a lot. I'm sure one unlike is purely unintentional.
Hi Hari, thanks for visiting and your kind remark :)
Best wishes to you!
Thanks a lot for sharing this
Very informative and useful. Thank you!
i don't understand a thing. i'm lost...
Thank you so much for this video! I was wondering: you do create a dummy variable for each characteristic. When running the regression, would I need to use one less dummy variable than I have? To avoid multicollinearity? That's what I understood from reviewing info on OLS models. Would so much appreciate your answer!
Did you get the answer?
Thanks a lot for the video. Maybe I understood something wrong. Why can you just use the tools for a normal multiple regression for panel data without controlling for the time or that the different measurements over time for each state do probably have something in comen?
VERY INTERESTING WITH SPSS. DIDNT KNOW SPSS IS THIS RICH!!!!
thanks for the nice illustration!!
really we appreciate your great efforts & best
thank you so much Hossam! best wishes!
Very much appreciated!
I appreciate you big time!
Thanks for this great video. Is it possible to use a stepwise approach to adding variables in blocks/steps where the dummy-coded variables would enter on their own? If so, would they enter first, last, or in some other temporal fashion in alignment with our theoretical framework/literature that guides the research?
In my case, I'm using undergraduate student-level data, and I'm trying to account for differences across institutions. I have 16 institutions, and I'm selecting one as a reference group to leave out of the regression.
Hi Kaitlyn,
If I am reading your question correctly, you are asking if you an enter the dummy variables first in an initial step in the first block, and then in the second block of variables set this to stepwise (to let the algorithm select the best set of within-case predictors). I have never done this myself or seen anyone else use this particular strategy before (truthfully, I am not a big fan of using stepwise approaches AND I have not really spent much time reflecting on this particular question before). As far as I can tell, there is nothing that would seemingly prohibit you from being able to do what you are describing in SPSS (I was able to accomplish it using the dataset from this video). In other words, Step 1 I entered the dummy variable with the entry method to ENTER; in Step 2, I entered the other predictors and reset the entry method to STEPWISE. Please be clear though, I am not endorsing or recommending this approach, but just telling you how it might be accomplished using SPSS.
I hope this helps! Cheers!
Great, thanks for your clear and thoughtful response! @@mikecrowson2462
Thank you for this helpful video! Is this same approach also possible using multivariate logistic regression (in SPSS, via complex sample package)?
hi thanks for the video. I have regression with control variables and fixed effects (country and time specific), how do i apply this in spss?
Thanks, reaaly appreciate it! I have a question though. If you have a (time varying control variable) how is the best way to use it in this model ?
Thanks for this video and your other great videos, Mike. Are you able to give any pointer on how to control for between-group variation using fixed effects dummy variables when the number of categories of the would-be dummy variable is very high? For example, country-level fixed effects when the number of countries is around 200. How might you create such a dummy variable in SPSS? Thanks again.
Suppose I have 500 cases measured over and over. Is it possible to create dummies for all these and use them in the model? I'm wondering how this model will look like!! Thank you
Thank's a lot! Can you tell something about the non-stationarity of the variables? Is it possible to ignore it using fixed effects?
Great video.
How does SPSS know that this is a panel data? On other Software we declare that this is panel data, but on SPSS I don't see that option.
Hi there. And thanks for the question. When you declare data to be panel data (such as in Stata if you are running a fixed-effects regression), the program takes care of various pre-processing steps prior to running the analysis. In theory, there are two different pre-processing steps can be used with our data when running a fixed-effects regression: Mean-deviation centering method or the creation of dummy coded variables (both methods assume your data are in long/vertical format). The default in Stata (I'll get to SPSS soon) when you use xtset along with the xtreg command and fe option is to recognize that panel data is being analyzed. As a result, Stata takes care of the pre-processing for you (using mean-deviation approach; see Allison, 2009) - and adjusts standard errors in your regression accordingly. If you are using the reg command in Stata, then the program would not recognize the data as panel. In this case, you would have to create dummy variables and then enter them into your regression (or, if you use the i. prefix, this can be used with your factor variable (which will lead to the automatic inclusion of the dummy variables in your regression). The latter method (Least-squares dummy variable approach) is essentially what we are doing in the example in my presentation with SPSS.
BOTH SPSS and Stata have the capability to use GEE (Generalized Estimating Equations) with longitudinal data (in fact, I have a video on it in Stata here: czcams.com/video/4vQfuIRIahY/video.html). In Stata, you would be using the xtreg command with pa (for population-averaged), whereas in SPSS you would be going through Generalized linear models --> Generalized estimating equations.
I am unclear as to whether you are wondering about SAS or R; but the logic I use above is the same. I can't speak much to SAS, but I do believe the 'plm' package for R uses the same approach (mean-deviations approach) as I noted above for SPSS.
I hope this answers your question!
@@mikecrowson2462 Yes, thank you.
This is great!!!
Could you reupload the dataset on google? I get an error 404 from google drive.
Brilliant :)
The R-Squared value for the panel regression in STATA is 0.3920 overall, which is lower than 0.819, right?
The R-square overall in Stata is computed differently than the R-square in spss, and I don't find the Stata overall R-square particularly useful. But I just created a new presentation on panel regression in spss with repeated measures data you might find helpful at czcams.com/video/cN_4RoeKlT4/video.html
Hello ! Very interesting video! I am a phd student and would like to know if I can contact you personally via email to ask a question concerning Fixed Effects models. Thank you
so nice and straight to the point. how does one chose between fixed effect and random effect using spss. how do does one know which model is best fit between fixed effect and random effect using spss in order to conduct a panel data analysis?
I am wondering the same. How to do Hausman test of specification in SPSS?
I spoilt so long organising data !!! You just taught in few clicks!!! I would have watched before!!! 😥
How is this model different from a multiple linear regression model?
Hi Haiying, it actually isn't different at all mathematically. It's basically a multiple regression using dummy coding of a group membership variable to control for any of the association between the iv's and dv that may be occurring between groups. By controlling for the variation occurring between groups, you end up with the within-groups associations remaining. Here is a link to a presentation that you might find helpful (of course the discussion is mainly in terms of Stata, but it does provides useful conceptual information): www.princeton.edu/~otorres/Panel101R.pdf
@@mikecrowson2462 Thank you very much professor! Does this mean that if you want to estimate time-fixed effects, you can also just simply add the time variable to the multiple linear regression (e.g., year) which would be just the same as using a time-fixed effects model?
is it possible to do hierarchical models with fixed effects?
The fixed effects model does not assume random intercepts or slopes as you would see in hierarchical models. And it does not allow you to model between-group predictors of the outcome as you could do in the case of a random effects model. SPSS does not have a random effects mechanism as far as I've found. Assuming a reasonably large number of groups randomly sampled from the population of groups (level 2 units in multilevel terminology), I generally prefer to use a multilevel modeling approach. But this fixed effects approach is pretty good when you have a smaller number of level 2 units. If you have predictors you wish to model at the group level (level 2), it is possible to include them only by modeling interactions involving level 1 predictors. Here's a nice article I found that you might find useful: biol609.github.io/Readings/McNeish_Kelley_PsychMethods_2019.pdf
Please upload video about fixed effect panel regression using first difference estimator
Please upload fixed effect regression using first difference estimator
thanks for this video sir i need your help kindly contact with me by mail