I have conducted research with young people where they completed several measures.
As the survey went on participants started to drop out so for example more people completed the wellbeing measure at the start of the study and less took part in the mental toughness questionnaire at the end of the study. This also happened with some sub scales, so for example the ERQ which has two sub scales, one sub scale had more participants than the other. My question is when reporting my sample size can I only include those participants who took part in all of the measures or can I report different numbers of PPS for each measure? I will be doing a multiple regression with the DV being the wellbeing measure (which most people took part in) and the IV's being the other measures which all had varying numbers of pps.
I would consider including individuals with missing scores to avoid bias and/or loss of statistical power. You can use either full information maximum likelihood (FIML) estimation or multiple imputation (MI). Both of these methods allow you to include all available data points. FIML is somewhat more straightforward to apply because it does not require imputations of missing values. Programs for structural equation modeling (e.g., AMOS, Mplus, lavaan) allow you to run standard multiple regression with FIML. MI is available in many programs as well.
See my Youtube videos on FIML & MI here:
FIML in Mplus: https://youtu.be/NMYHADPhM2U
https://youtu.be/x61bDYpSvbE
MI in Mplus: https://youtu.be/EmXG2ZGw0GE
Christian Geiser Thank you so much! I've just been watching your videos and they are incredible helpful because I am also wanting to use structural equation modelling to see which of my IV's mediate my DV after the regression so you have a new subscriber!
Awesome, I'm so glad my Youtube content is helpful for you. Best of luck with your work!
Also, I forgot to mention that JASP is a free point-and-click software that allows you to easily run regression and mediation/path models based on the R package lavaan. The lavaan package uses FIML by default, so missing values are automatically accounted for. See my Youtube video on how to use JASP/lavaan for path/mediation analysis:
https://youtu.be/uVZryVecL8s
I also have a playlist on statistical analysis in JASP that you can check out for additional techniques including CFA & SEM with latent variables:
https://www.youtube.com/playlist?list=PL-kVjeOVYChrZfRkP6R0RRW0uV0cvaZB-
Hi, there is a rule that missing data per observation or indicator should not exceed 15%. If missing data is less than 15% we can use all the data, if not the observation or the indicator must be deleted. You should also evaluate the possibility to use missing data replacement. There are several methods that can be used, such us mean value replacement, pairwise deletion and methods based on regression.
- Badges
- Science topic