Dear Crowson, I researched to determine the mediating role of self-efficacy in the relationship between two thinking skills. Since this research was cross-sectional, I received criticism that it would not be appropriate to conduct a mediator variable analysis. In this case, I decided to design a longitudinal study to test my hypotheses. I administered my survey to the same sample three times. (i) How long should there be between applications? (ii) I want to test the mediating role of self-efficacy using latent growth curve mediation model analysis. For this, should I create intercept and slope variables for my dependent, independent, and mediator variables? (iii) Is there a different method you can suggest?
Hi Mike, you model @ 10:02 is wonderful in this presentation. You used a mediator (math grade) at time one. But the mediator in my study measured 3 time through the study. How can we fit all three time points in the above model?
Thanks for clarifying your question. I think I have a better understanding of what you are asking. First, I'd like to mention that the actual dataset did contain math grades at 3 time points (that corresponded with the math ability measurements). However, since the demonstration was designed to test for a mediated effect at Level 2 (in conventional HLM terminology) - i.e., I was attempting to predict variation between persons in slopes and intercepts as a function of sex and a "math achievement" variable - I just utilized the time 1 math grades variable as a Level 2 mediator (treating it as a between-subjects variable, rather than something that varies over time). It sounds like you may be asking about mediation involving two growth curves (e.g., math ability and math grades), each having their own intercept and slope parameters that vary between individuals at Level 2. You can certainly test for mediation involving growth curve factors (i.e., slope and intercept factors) from two (or more) domains. To do this, you'd need to create a dual domain model (see my other video on testing growth curves in dual domains), and then add in a level 2 predictor (e.g., sex; treatment, etc.) with arrows pointing from that variable to the "mediating" growth curve component(s)/factor(s) associated with one domain. Additionally, you would have arrows pointing from the mediating growth curve factors to the growth factors from the second domain. For instance, let's say I wished to test whether the relationship between sex and change (over time) in perceived math ability is mediated by changes in students' math grades/performance. I would first create a model with slope and intercept factors for "grades" and "perceived ability". Thus, this model would include 4 factors (2 slope factors and 2 intercept factors). Next, I would include sex as a predictor of the mediator (i.e., the slope factor associated with "grades"). The slope factor for "grades" in turn would be treated as a predictor of the slope factor for "perceived ability". In short, I'd draw and arrow from sex to the slope factor for grades and an arrow from the slope for grades to the slope factor for perceived ability. If I thought the mediated effect may be a partial (instead of full one), then I could also theoretically draw a direct effect from sex to the slope factor for perceived ability. In effect, this model is testing whether the effect of sex on individuals' change over time in math ability is mediated by change over time in their math grades. You can use the same logic in the context of testing mediated treatment effects on change over time in an outcome measure. Let's say that instead of using sex as our predictor (at level 2), we use treatment as our between-subjects factor. Theoretically, we might assume that the effect of the treatment on the rate of change in perceived math ability should be mediated via its impact on changes in actual performance (i.e., reflected in grades). We can test this out using the same process described above, with the resulting model assuming that treatment impacts changes in grades, which then impacts changes in perceived ability. I hope this is helpful to you.
Dear Crowson, I researched to determine the mediating role of self-efficacy in the relationship between two thinking skills. Since this research was cross-sectional, I received criticism that it would not be appropriate to conduct a mediator variable analysis. In this case, I decided to design a longitudinal study to test my hypotheses. I administered my survey to the same sample three times. (i) How long should there be between applications? (ii) I want to test the mediating role of self-efficacy using latent growth curve mediation model analysis. For this, should I create intercept and slope variables for my dependent, independent, and mediator variables? (iii) Is there a different method you can suggest?
for mediator with multiple time point, what doe the model looks like? how can we compare two treatment arms?
Hi there. I'm a bit unclear about the model you are asking about.
Hi Mike, you model @ 10:02 is wonderful in this presentation. You used a mediator (math grade) at time one. But the mediator in my study measured 3 time through the study. How can we fit all three time points in the above model?
Thanks for clarifying your question. I think I have a better understanding of what you are asking. First, I'd like to mention that the actual dataset did contain math grades at 3 time points (that corresponded with the math ability measurements). However, since the demonstration was designed to test for a mediated effect at Level 2 (in conventional HLM terminology) - i.e., I was attempting to predict variation between persons in slopes and intercepts as a function of sex and a "math achievement" variable - I just utilized the time 1 math grades variable as a Level 2 mediator (treating it as a between-subjects variable, rather than something that varies over time). It sounds like you may be asking about mediation involving two growth curves (e.g., math ability and math grades), each having their own intercept and slope parameters that vary between individuals at Level 2. You can certainly test for mediation involving growth curve factors (i.e., slope and intercept factors) from two (or more) domains. To do this, you'd need to create a dual domain model (see my other video on testing growth curves in dual domains), and then add in a level 2 predictor (e.g., sex; treatment, etc.) with arrows pointing from that variable to the "mediating" growth curve component(s)/factor(s) associated with one domain. Additionally, you would have arrows pointing from the mediating growth curve factors to the growth factors from the second domain. For instance, let's say I wished to test whether the relationship between sex and change (over time) in perceived math ability is mediated by changes in students' math grades/performance. I would first create a model with slope and intercept factors for "grades" and "perceived ability". Thus, this model would include 4 factors (2 slope factors and 2 intercept factors). Next, I would include sex as a predictor of the mediator (i.e., the slope factor associated with "grades"). The slope factor for "grades" in turn would be treated as a predictor of the slope factor for "perceived ability". In short, I'd draw and arrow from sex to the slope factor for grades and an arrow from the slope for grades to the slope factor for perceived ability. If I thought the mediated effect may be a partial (instead of full one), then I could also theoretically draw a direct effect from sex to the slope factor for perceived ability. In effect, this model is testing whether the effect of sex on individuals' change over time in math ability is mediated by change over time in their math grades. You can use the same logic in the context of testing mediated treatment effects on change over time in an outcome measure. Let's say that instead of using sex as our predictor (at level 2), we use treatment as our between-subjects factor. Theoretically, we might assume that the effect of the treatment on the rate of change in perceived math ability should be mediated via its impact on changes in actual performance (i.e., reflected in grades). We can test this out using the same process described above, with the resulting model assuming that treatment impacts changes in grades, which then impacts changes in perceived ability. I hope this is helpful to you.
@@mikecrowson2462 Thank you. This was a helpful explanation of what I needed. Also, great vids