I would essentially like to explore the following: (H1) the number of firms participating in program X in a region on firm performance in program X; (H2) the distance between a firm and the nearest Y on firm performance in program X; and (H3) the influence of the number of firms participating in program X in a region on the relationship between the distance between a firm and the nearest Y and firm performance in program X (i.e. a moderation effect).
I have data for 88 firms that participated in program X over 8 years (I also have data for hundreds and hundreds of firms that did not participate, however, I'm not currently using that data). The "average" firm participated in program X for 4.4 years (unbalanced panel).
Here are the variables I'd like to include in the model:
DV = ordinal measure of firm performance in program X (1 - 5 scale, 1 = lowest performance and 5 = highest performance)
IV1 = continuous, time-variant measure of the number of participating firms in a region
IV2 = continuous, time-invariant measure of the distance to nearest Y
IV3 = continuous, time-invariant measure of the distance to nearest Y-squared
Moderation = IV1 * IV2 (number of participating firms in a region will moderate the effect of distance to nearest Y on firm performance)
Controls = firm reputation (categorical/continuous variable), firm size (continuous), and firm location
***
I'd like to point a few things out. First, I'm a graduate student and have a great deal to learn (resource suggestions and as lay terms as possible will be greatly appreciated). Second, one of the reasons I'm at a loss as to which model to select is that despite the fact that I have both time variant and time invariant variables, which alone would generally lead me to believe I ought to choose a random effects ordinal logit/probit (xtologit/xtoprobit in stata), the characteristics of my RQ and theory might indicate a fixed effects model. My research question is essentially - how will place/location influence firm performance in program X? I obviously can't measure 'place' (at least not in the way I'd like - 'place' is far too complex) and have thus come up with two aspects of place to determine whether or not 'place' (or at least these two aspects of place) influence firm performance in program X. Thus, there are definitely omitted variables and the omitted variables are likely correlated with the IVs included in the model (for example, I can't measure manager's attitudes but manager's attitudes are almost certainly correlated with my second IV). Consequently, I wonder if a fixed effects model would be better. Yet, I'm not exactly sure how to run a FE model - perhaps ologit/oprobit y x1 x2 x3, vce(cluster year) - or, y x1 x2 x3 i.year, vce(cluster id), or something like that? Anyway, still trying to figure all of this out in terms of this RQ, theory, and this data.
Any help would be greatly appreciated. Thank you!!!!
P.S. I initially assumed the RE would be appropriate and posted a question about the results here - http://www.statalist.org/forums/foru...-with-xtologit. However, after thinking about this model quite intensely the past several days, I've made a few changes and those results have (obviously) been thrown out. I'm trying to take a step back and figure out which model best fits the RQ, theory, and data... Thanks again!
I have data for 88 firms that participated in program X over 8 years (I also have data for hundreds and hundreds of firms that did not participate, however, I'm not currently using that data). The "average" firm participated in program X for 4.4 years (unbalanced panel).
Here are the variables I'd like to include in the model:
DV = ordinal measure of firm performance in program X (1 - 5 scale, 1 = lowest performance and 5 = highest performance)
IV1 = continuous, time-variant measure of the number of participating firms in a region
IV2 = continuous, time-invariant measure of the distance to nearest Y
IV3 = continuous, time-invariant measure of the distance to nearest Y-squared
Moderation = IV1 * IV2 (number of participating firms in a region will moderate the effect of distance to nearest Y on firm performance)
Controls = firm reputation (categorical/continuous variable), firm size (continuous), and firm location
***
I'd like to point a few things out. First, I'm a graduate student and have a great deal to learn (resource suggestions and as lay terms as possible will be greatly appreciated). Second, one of the reasons I'm at a loss as to which model to select is that despite the fact that I have both time variant and time invariant variables, which alone would generally lead me to believe I ought to choose a random effects ordinal logit/probit (xtologit/xtoprobit in stata), the characteristics of my RQ and theory might indicate a fixed effects model. My research question is essentially - how will place/location influence firm performance in program X? I obviously can't measure 'place' (at least not in the way I'd like - 'place' is far too complex) and have thus come up with two aspects of place to determine whether or not 'place' (or at least these two aspects of place) influence firm performance in program X. Thus, there are definitely omitted variables and the omitted variables are likely correlated with the IVs included in the model (for example, I can't measure manager's attitudes but manager's attitudes are almost certainly correlated with my second IV). Consequently, I wonder if a fixed effects model would be better. Yet, I'm not exactly sure how to run a FE model - perhaps ologit/oprobit y x1 x2 x3, vce(cluster year) - or, y x1 x2 x3 i.year, vce(cluster id), or something like that? Anyway, still trying to figure all of this out in terms of this RQ, theory, and this data.
Any help would be greatly appreciated. Thank you!!!!
P.S. I initially assumed the RE would be appropriate and posted a question about the results here - http://www.statalist.org/forums/foru...-with-xtologit. However, after thinking about this model quite intensely the past several days, I've made a few changes and those results have (obviously) been thrown out. I'm trying to take a step back and figure out which model best fits the RQ, theory, and data... Thanks again!