Growth Curve Episode 2: The Coding Of Time

[@cdee3346]

You are the most talented and passionate teacher I have seen! Thanks for sharing your knowledge!

[@centerstat]

Hi Cici -- thanks for your very sweet words. I really do appreciate it. I hope you find these of some use -- good luck in your work. Take care -- patrick

[@luyaoliang650]

you are an amazing teacher! thank you so much!

[@a45701]

Great video. By the way, how did you get the camera stay in focus on the whiteboard? What type of camera did you use for filming this?

[@centerstat]

Thanks for the comment. It took me forever to figure out the focus -- I just use an old Canon G9X point-and-shoot that also takes HD video. I drew an "x" on the board, set the focal length, and turned the auto-focus off. Nothing like a low-tech solution.

[@a45701]

​@@centerstat Thanks, interesting. I tried using a whiteboard when teaching a remote class, but haven't found a webcam that works. Thanks for putting up those videos, also setting examples on what is an effective teaching and presentation style.

[@michellelee2700]

Super helpful explanation of the foundations. Thank you so much for your insightful teaching!

[@mirjamsnel-deboer431]

Hi Patrick, thank you for your teaching, it has helped me very much. I have a question. I have data from 1 week with 7 time points, the respondents filled out a questionnaire every day at approximately the same time. I think it would make sense to start with zero and end with 6, but I read somewhere that you can also start with zero and end with 1 and use .17 .33 .5 and so on in between. Why are the estimates different (you spoke about that in your video) and how do I know which to use?

[@centerstat]

Hi Mirjam -- thanks for your nice note. You raise a great question. If you fix the values of time to pre-defined values (so don't freely estimate them), then it makes no difference how you code them as long as you are cognizant of what a "one unit change in time" means. So you can start with zero, have zero in the middle, or end with zero. Or as you say, you can increment say 0, 1, 2, 3, or equivalently 0, .333, .666, 1 -- these are all equivalent in all ways except the slope factor will be scaled by whatever "one unit of change" represents. That is, is a one unit change a day? a week? a month? A buddy of mine wrote a great paper on this if you're interested -- the cite is below. Hope this helps -- take care -- patrick Biesanz, J. C., Deeb-Sossa, N., Papadakis, A. A., Bollen, K. A., & Curran, P. J. (2004). The role of coding time in estimating and interpreting growth curve models. Psychological Methods, 9, 30-52.

[@aungmyohtut2440]

That's a very clear explanation. Thanks so much!

[@sharonanderson7657]

Wow, What a clear, understandable presentation.

[@saratodorovic9273]

Great video! I was wondering if it is feasible or even possible to use only two time points for a latent class growth analysis?

[@centerstat]

Hi Sara -- thanks for the nice comment. Briefly, you need at least three time points to over-identify a straight line, so that is usually taken as the minimum needed for growth modeling whether it be in a single group or mixture setting. You could do a more standard residualized--change model in which you regress y2 on y1 as well as other predictors of y2 -- so you are looking at predictors of y2 above and beyond the effects of y1. That may or may not work in your situation -- good luck with your work -- patrick

[@omerrr09]

Hi professor, i have data with 3 time points. Change from time 0 to time 1 is much more than to time 1 to time 2. So i analysed the data with free estimation instead of traditionall time coding as 0,1, and 2. Free estimation provided perfect fit indices, while fixed time coding worse indices. It reasobable because if i understand truly, time coding is functioning as the determinant of change units between time points. So if we use fixed time coding, we assume that there is steady and almost same amounts of change across time. However, in real world it is rarely seen. Then why we use fixed time points? Thanks in advance

[@centerstat]

Thanks for your question. The reason you're getting perfect fit is that the model is saturated -- that is, you are estimating as many parameters as pieces of information you observed, and thus you have zero degrees-of-freedom. Unfortunately, with 3 time points, a linear trajectory is all you can estimate (there are ways you can trick the model into a quadratic, but we don't recommend this because it will fit your data perfectly). Another option is to abandon the growth model altogether and move to something like a latent change score model or auto-regressive cross-lagged models, but those too have pros and cons. Sorry I can't be more helpful -- patrick

[@omerrr09]

@@centerstat Dear Patrick, thanks for your answer. In fact, i am novice at this analysis method, so i will keep informations you provided in my mind.

[@shangshang3429]

Thank you for your time. it is really helpful and useful.

[@majidghasemy7482]

Thanks a lot for this great video.

[@taehyunalisonlee6766]

NO, THANK YOU FOR YOUR TIME!

[@장동일-b6s]

Wonderful video. Thank you.

[@taniachavez2122]

Is it wrong to set the time as a factor variable?

[@centerstat]

Hi Tania -- thanks for your question. I'm not sure what exactly you mean by setting time as a factor variable? In the SEM approach, the numerical values of time are fixed values for factor loadings that in turn define the underlying growth factors -- intercept, slope, curvature, etc. So in that sense, yes -- time can be set as a latent factor, but only in as much as they define the underlying growth trajectories. I hope this helps -- good luck with your work -- patrick

[@qiqqqqq7158]

Amazing! Thank you!

[@qiujieli5930]

Very clear!

[@valentinagiaconi4508]

Thank you very much!