Hello, i want to use structural equations modelling given that i'm interested in the effect of a latent variable "disclosure" on firm value, i have questions about how to model this relationship with longitudinal or panel data, shoul i just use cluster by firm or what sem model shoul i use ? i have tree indicators for my latent variable.
Hi Luisa,
it is not really clear how your model looks like? You have 1 latent with 3 indicators as IV and one DV? In addition, how many waves do you have?
Basically, when you have 2 waves, a cross-lagged panel model would be possible that allows testing lagged or synchronous effects in both directions (IV ->DV or reverse), see
Hamaker, E. L., Kuiper, R. M., & Grasman, R. P. P. P. (2015). A critique of the cross-lagged panel model. Psychological Methods, 20(1), 102-116.
With 3 or more waves, you can again test a panel model or alternatively, a growth curve model. If I remember correctly, the ALT model can integrate both when you have 4 waves, see
Bollen, K. A., & Curran, P. J. (2004). Autoregressive latent trajectory (ALT) models A synthesis of two traditions. Sociological Methods & Research, 32(3), 336-383. doi:10.1177/0049124103260222
Beyond that here are some introductions to longitudinal analysis:
Goldsmith, K. A., MacKinnon, D. P., Chalder, T., White, P. D., Sharpe, M., & Pickles, A. (2018). Tutorial: The practical application of longitudinal structural equation mediation models in clinical trials. Psychological Methods, 23(2), 191-207. doi:10.1037/met0000154
Newsom, J. T. (2015). Longitudinal structural equation modeling: A comprehensive introduction. New York: Routledge.
Ployhart, R. E., & Vandenberg, R. J. (2010). Longitudinal research: The theory, design, and analysis of change. Journal of Management, 36(1), 94-120.
HTH
Holger
Thanks Holger, i am new on this SEM modelling. i just updated mi question with a graphic for better understanding. firmsize leverage and growth are the control variables for the regression of disclosure on tobinq(firm value), and sim readab and unc are the measured variables for the latent disclosure.
Hi Luisa,
ok, than I guessed right. I would start with a test of the latent structure by means of a CFA. You could do that by starting by separating the waves (to keep it simple) or in the multi-wave model. In the latter you have the structure as in your diagram but
a) all variables within and across waves correlate with each other (instead of having effects)
b) the errors of indicators correlate with each other across time.
If this is unclear, the papers I referred to should tell you more.
Best,
Holger
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling