Hello Everyone,
We are conducting research that is exploring the difference in recovery trajectories for persons with schizophrenia and persons with bipolar disorder. Over two years, we gathered longitudinal data and then analyzed those differences using -mixed- (using Stata 13). As we hypothesized, there was a main effect for group differences (i.schzbip = schizophrenia/bipolar disorder). As part of the analyses we ran post-estimation follow-up with a -margins- command (syntax below).
mixed gas age-sedhyp c.time alcab canab i.schzbip || _all: r.time || id:
margins i.schzbip, atmeans
--------------------------------------------------------------------------------
| Delta-method
| Margin Std. Err. z P>|z| [95% Conf. Interval]
---------------+----------------------------------------------------------------
schzbip |
schizophrenia | 51.60243 3.198098 16.14 0.000 45.33427 57.87059
Bipolar 1 & 2 | 58.45254 3.3876 17.25 0.000 51.81296 65.09211
--------------------------------------------------------------------------------
I want to describe the magnitude of the difference between the groups (i.e., effect size), and started computing Cohen's d. However, when I started going through the process I realized that I wasn't calculating the raw differences between groups, I was calculating group differences while many other variables were being controlled for. This raised several questions for me:
1. Is Cohen's d still valid as an index of effect size when controlling for other variables?
2. If yes, how does one determine SD post-estimation? Converting SE to SD is straightforward if you have the N, but Stata uses a computation that produces the delta-method SE, and despite reading over it's calculation, I'm unsure of how I can get SD from it.
3. On a slightly different tangent, why does Stata provide z/p values for the marginal mean estimates? I understand how they're calculated (margin/SE = z), and I'm guessing Stata is providing them for a reason, but I can't figure out what that reason is.
Thanks a lot,
David.
Please note that Elc Estrera asked a variation of this question previously in an untagged post, but there wasn't any feedback on it.
We are conducting research that is exploring the difference in recovery trajectories for persons with schizophrenia and persons with bipolar disorder. Over two years, we gathered longitudinal data and then analyzed those differences using -mixed- (using Stata 13). As we hypothesized, there was a main effect for group differences (i.schzbip = schizophrenia/bipolar disorder). As part of the analyses we ran post-estimation follow-up with a -margins- command (syntax below).
mixed gas age-sedhyp c.time alcab canab i.schzbip || _all: r.time || id:
margins i.schzbip, atmeans
--------------------------------------------------------------------------------
| Delta-method
| Margin Std. Err. z P>|z| [95% Conf. Interval]
---------------+----------------------------------------------------------------
schzbip |
schizophrenia | 51.60243 3.198098 16.14 0.000 45.33427 57.87059
Bipolar 1 & 2 | 58.45254 3.3876 17.25 0.000 51.81296 65.09211
--------------------------------------------------------------------------------
I want to describe the magnitude of the difference between the groups (i.e., effect size), and started computing Cohen's d. However, when I started going through the process I realized that I wasn't calculating the raw differences between groups, I was calculating group differences while many other variables were being controlled for. This raised several questions for me:
1. Is Cohen's d still valid as an index of effect size when controlling for other variables?
2. If yes, how does one determine SD post-estimation? Converting SE to SD is straightforward if you have the N, but Stata uses a computation that produces the delta-method SE, and despite reading over it's calculation, I'm unsure of how I can get SD from it.
3. On a slightly different tangent, why does Stata provide z/p values for the marginal mean estimates? I understand how they're calculated (margin/SE = z), and I'm guessing Stata is providing them for a reason, but I can't figure out what that reason is.
Thanks a lot,
David.
Please note that Elc Estrera asked a variation of this question previously in an untagged post, but there wasn't any feedback on it.
Comment