Hi folks, I made a few edits/corrections and additions to the Supplemental Powerpoint (last edited Sept 23) this morning. You should be able to download this updated version by following that link under the video description. Thanks!
Hi Mike, awesome video! For my thesis I am predicting the effect certain innovations have on operational performance. I have a similar data set in SPSS. However, my IVs are only measured in one point in time (Technology implemented in 2018). My DVs are measured in three points in time (2018,2019,2020). Would this model still be applicable? Kind regards! keep up the good work!
Question: does this dummy variable technique work when there is a large sample size? If n is greater than 100, I could see this becoming a cumbersome strategy. Thanks for the informative video.
Mike -- super useful! I've got 1190 participants after excluding any missing data, and it seems to really balk -- it'll only show 2 of my 4 blocks of analysis, and will say MXCELLS exceeds the limits, and SPSS28 doesn't allow for adjustment of MXCELLS any more. I'm wondering if you have any advice?
Thank you! Is using this method (linear regression with these dummies) the same as using the mixed models analysis using the subjects and repeated measures options?
Dear Professor Mike Crowson, I would like to know exactly the reason to carry out an approach like this and not a GHG, for example. What are the differences and advantages? Thanks for the classes!
Hi Mike, This video is very helpful, however, my sample size is 4000, and it is no feasible to create a dummy variable for each “id”. Could you please provide me with alternative method?
Thanks for this video, Mike. Quick question: if I wanted to incorporate time fixed effects into the model, could I just enter the 'Index' nominal variable into the 'create dummy variable' function and create n-1 dummies for time, since this variable measures temporal occasions of measurement?
Hi David, that is absolutely correct. In fact, yesterday I put together a video where I was showing how it is possible to reparameterize a one-way ANOVA using the dummy variable approach for both individual and time: ruclips.net/video/4P_H4JNOLL8/видео.html Cheers!
Hi David, one other thing. You can use the dummy variable approach with the time factor. But another possibility in those cases where you are interested in modeling the shape of change (apart from linear) might be to create a time variable (coded 0, 1, 2, 3,... for each time point) and then add higher order polynomial terms to the model. This approach is also demonstrated in the context of a multilevel modeling analysis in www.taylorfrancis.com/books/mono/10.4324/9780203701249/multilevel-longitudinal-modeling-ibm-spss-ronald-heck-scott-thomas-lynn-tabata .
Thanks Mike, What if our latent variables are described by different items. Shall we include the mean scores of each variable and treat it as a latent variable in this approach?
Dear Mike, I have content analysis of 100 movie characters and survey data (Likert scale) of 565 respondents. What statistical tool or method should I use to correlate both data sets?
Hi Mike. what if my data set is cross sectional time series with change in policy (treatment) in the middle. I have 100 organisations, each has 6 years observation of 4 variables. 3 years pre policy change, 3 years post policy change. I am including one dummy variable for change in policy (0,1), is this correct? second how we can run the Hausman test in SPSS to check of fixed module or random model is suitable?
Hi Chalobol, I'm not sure what you mean by p (unless it simply is just notation for a trend component). Generally, you focus on the issue of trend if that is a substantive research question of interest to you. Some folks want to study growth (or decay) over time on some variable, and that is generally when one includes predictors (in the form of time indicators) in a model to test the shape of change over time. [In fact, you see this in the context of growth curve modeling.] If that is purely your interest, then you probably would not be including other time-varying predictors (such as those I cover in this video). On the other hand, there may be situations where you wish to include time-varying predictors like the ones in the video while also attempting to see if a growth curve trajectory might remain (as well as its nature) after including those time-varying predictors in your model. There are a couple of ways to approach adding time-indicators in the model. One is to (assuming you have enough time points to treat it as reasonably continuous) modify your regression to include additional variables (e.g., time, time-squared, time-cubed) to capture linear, quadratic, or cubic (generally we don't go much higher than this) trends in your data. [You might do some reading on polynomial regression.] Another option, which is probably better when you have very few time points is to include a time-factor, but to use special coding of the binary predictors (representing time) in your model to test out various trends. I really don't do much with the latter. However, I believe Cohen, Cohen, West, and Aiken (2003) discuss this type of coding in their text. By the way, you might also check out the chapters on modeling longitudinal/repeated measures data in Heck, Thomas, and Tabata's (2014) book on multilevel modeling in SPSS (see citations below). Although done in the context of multilevel modeling, they do address coding of longitudinal data using both approaches I mentioned above. I hope this helps! Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed.). Mahwah, NJ: Lawrence Erlbaum. Heck, R. H., Thomas, S. L., & Tabata, L. N. (2014). (2nd ed.). Routledge/Taylor & Francis Group.
@@mikecrowson2462 actually, I don't really know what is p for trend. I saw it in the article which was the longitudinal study. So, I don't know what difference between p-value and p( value)for trend?
Hi folks, I made a few edits/corrections and additions to the Supplemental Powerpoint (last edited Sept 23) this morning. You should be able to download this updated version by following that link under the video description. Thanks!
Thank you so much for your detail and careful guidance!
You just saved my master thesis. Thank you
Hi Mike, awesome video! For my thesis I am predicting the effect certain innovations have on operational performance. I have a similar data set in SPSS. However, my IVs are only measured in one point in time (Technology implemented in 2018). My DVs are measured in three points in time (2018,2019,2020). Would this model still be applicable? Kind regards! keep up the good work!
I have this situation as well. Did you ever figure out the answer?
Question: does this dummy variable technique work when there is a large sample size? If n is greater than 100, I could see this becoming a cumbersome strategy. Thanks for the informative video.
Mike -- super useful! I've got 1190 participants after excluding any missing data, and it seems to really balk -- it'll only show 2 of my 4 blocks of analysis, and will say MXCELLS exceeds the limits, and SPSS28 doesn't allow for adjustment of MXCELLS any more. I'm wondering if you have any advice?
Thank you! Is using this method (linear regression with these dummies) the same as using the mixed models analysis using the subjects and repeated measures options?
Thank you.
Dear Professor Mike Crowson,
I would like to know exactly the reason to carry out an approach like this and not a GHG, for example. What are the differences and advantages? Thanks for the classes!
thanks mike!
Hi Mike,
This video is very helpful, however, my sample size is 4000, and it is no feasible to create a dummy variable for each “id”.
Could you please provide me with alternative method?
Thanks for this video, Mike. Quick question: if I wanted to incorporate time fixed effects into the model, could I just enter the 'Index' nominal variable into the 'create dummy variable' function and create n-1 dummies for time, since this variable measures temporal occasions of measurement?
Hi David, that is absolutely correct. In fact, yesterday I put together a video where I was showing how it is possible to reparameterize a one-way ANOVA using the dummy variable approach for both individual and time: ruclips.net/video/4P_H4JNOLL8/видео.html
Cheers!
Hi David, one other thing. You can use the dummy variable approach with the time factor. But another possibility in those cases where you are interested in modeling the shape of change (apart from linear) might be to create a time variable (coded 0, 1, 2, 3,... for each time point) and then add higher order polynomial terms to the model. This approach is also demonstrated in the context of a multilevel modeling analysis in www.taylorfrancis.com/books/mono/10.4324/9780203701249/multilevel-longitudinal-modeling-ibm-spss-ronald-heck-scott-thomas-lynn-tabata .
@@mikecrowson2462 Great, thanks so much, I'll watch the other video.
Thanks Mike,
What if our latent variables are described by different items. Shall we include the mean scores of each variable and treat it as a latent variable in this approach?
Thank you Mike! Is it somehow possible to add control variables?
Dear Mike, I have content analysis of 100 movie characters and survey data (Likert scale) of 565 respondents. What statistical tool or method should I use to correlate both data sets?
Hi Mike. what if my data set is cross sectional time series with change in policy (treatment) in the middle. I have 100 organisations, each has 6 years observation of 4 variables. 3 years pre policy change, 3 years post policy change. I am including one dummy variable for change in policy (0,1), is this correct? second how we can run the Hausman test in SPSS to check of fixed module or random model is suitable?
Thank you very much.
Could you explain about p for trend? I wonder when and how should I use p for trend.
Hi Chalobol, I'm not sure what you mean by p (unless it simply is just notation for a trend component). Generally, you focus on the issue of trend if that is a substantive research question of interest to you. Some folks want to study growth (or decay) over time on some variable, and that is generally when one includes predictors (in the form of time indicators) in a model to test the shape of change over time. [In fact, you see this in the context of growth curve modeling.] If that is purely your interest, then you probably would not be including other time-varying predictors (such as those I cover in this video). On the other hand, there may be situations where you wish to include time-varying predictors like the ones in the video while also attempting to see if a growth curve trajectory might remain (as well as its nature) after including those time-varying predictors in your model. There are a couple of ways to approach adding time-indicators in the model. One is to (assuming you have enough time points to treat it as reasonably continuous) modify your regression to include additional variables (e.g., time, time-squared, time-cubed) to capture linear, quadratic, or cubic (generally we don't go much higher than this) trends in your data. [You might do some reading on polynomial regression.] Another option, which is probably better when you have very few time points is to include a time-factor, but to use special coding of the binary predictors (representing time) in your model to test out various trends. I really don't do much with the latter. However, I believe Cohen, Cohen, West, and Aiken (2003) discuss this type of coding in their text. By the way, you might also check out the chapters on modeling longitudinal/repeated measures data in Heck, Thomas, and Tabata's (2014) book on multilevel modeling in SPSS (see citations below). Although done in the context of multilevel modeling, they do address coding of longitudinal data using both approaches I mentioned above. I hope this helps!
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed.). Mahwah, NJ: Lawrence Erlbaum.
Heck, R. H., Thomas, S. L., & Tabata, L. N. (2014). (2nd ed.). Routledge/Taylor & Francis Group.
@@mikecrowson2462 thank you so much.
@@mikecrowson2462 actually, I don't really know what is p for trend. I saw it in the article which was the longitudinal study. So, I don't know what difference between p-value and p( value)for trend?
am I the only one cannot see the video? something seems wrong?