|
|
Logistic Regression Montecarlo Analysis |
|
Message/Author |
|
|
Hi Drs. Muthen, In Example 3.5, the logistic regression code in the User's Guide examples, I am using the corresponding montecarlo code to run power for a logistic regression model. I am positing that my outcome variable, u1, has a frequency of the outcome of 78%. I think perhaps that can be indicated by where I set the threshold in the line, "[u1$1*1]" but I'm not sure what a threshold of 1 is. Is it 1 standard deviation above the mean? Or, is it on the -15 to +15 scale? Thanks so much for your help! Mary Mitchell |
|
|
You know that with logit, P(u=1 | x) = 1/(1 +exp(-L)), where L is the logit and L = threshold - b*x. So you can compute P from L with say x=0. |
|
|
Dear Dr. Muthen, Thank you for the formula! For a threshold of 1.3, I exponentiated -1.3 and got .27, which I plugged in and got a probability of .785, which is close to the percentage of respondents I needed in one of the categories (i.e., when x = 0). Mary |
|
|
Sorry, Logit = -threshold + b*x. But you did the right thing exponentiating -1.3. |
|
Back to top |
|
|