|
|
 |
| Cross Sectional Data with a Preponder... |
 
|
| Message/Author |
|
|
|
|
| I attended your session in Baltimore last week at Hopkins, and learned immensely. Thank you for a great presentation. I am trying to find an example in the text that fits my analysis needs. I am trying to see if there are latent groups with differing symptom experiences in a sample of patients with advanced melanoma. The data are individual items from a symptom checklist, where various symptoms are listed, and the respondent is asked to indicate the severity of the symptom on a scale of 0-4. 0 is if they don't have the symptom, and 4 is the worst severity possible. As you might expect, there are a preponderance of zeros in this data set, and so I have been looking at whether a 2 part model is the right approach. I can find some examples of a 2 part model for longitudinal data but I am hoping you can point me to a paper or examples in your text of this approach for a cross-sectional analysis. Thank you in advance for your help. Sandra |
|
|
|
|
Thanks for the kind words.
We are finishing up a paper on 2-part factor analysis and the steps to do this, so that's a cross-sectional analysis that might be helpful. It will be posted shortly.
However, unless you have a strong reason to use 2-part in this example, I think perhaps it more straightforward to treat your outcome as (ordered) categorical. The LCA approach for this is just like for binary outcomes.
2-part is more suitable when you have a more continuous tail for the non-zero part - you have only the categories 1, 2, 3, 4. Also, for 2-part you might want to have a hypothesis that the choice of being in zero vs not zero has different antecedents than the non-zero values.
If you treat your outcome as categorical you may want consider the need to collapse the 2 highest categories to not have too few observations there. |
|
|
| Yu-Chung Su posted on Saturday, April 18, 2020 - 4:41 am
|
|
|
HI Dr. Muthen,
Could you post the paper you refered to? (is it ablout using zero-inflated Poisson model?)
Thank you for you time
Yu-Chung |
|
|
|
|
The paper is available on our website:
Kim, Y.K. & Muthén, B. (2009). Two-part factor mixture modeling: Application to an aggressive behavior measurement instrument. Structural Equation Modeling, 16, 602-624.
Zero-inflated Poisson is discussed here in Section 6.5.1:
Muthén, B. & Asparouhov, T. (2009). Growth mixture modeling: Analysis with non-Gaussian random effects. In Fitzmaurice, G., Davidian, M., Verbeke, G. & Molenberghs, G. (eds.), Longitudinal Data Analysis, pp. 143-165. Boca Raton: Chapman & Hall/CRC Press.
download paper contact first author show abstract |
|
|
| Yu-Chung Su posted on Saturday, April 18, 2020 - 6:40 am
|
|
|
Thank you Dr.Muthen
I am currently analyzing LCA on baseline(bl) (first step of LTA), so USEVARIABLE will only include _bl
NPIstobl is the zero-inflated Poisson variable, I don't know the following syntax after COUNT = NPIstobl(i);
Could you show me how to do it? I still can't figure it out after reading EXAMPLE 7.11 in User Guide
VARIABLE:
NAMES ARE
id age sex edu dx_bl outcome interval
NPIstobl mem_bl ef_bl lan_bl ca_bl
NPIsto6m mem_12m ef_12m lan_12m ca_12m
;
USEVARIABLE ARE NPIstobl mem_bl ef_bl ca_bl
;
IDVARIABLE = id
;
MISSING ARE ALL (-999)
;
CLASSES = C (2)
;
ANALYSIS:
TYPE = mixture
;
ESTIMATOR = MLR
;
MODEL:
COUNT = NPIstobl(i); |
|
|
|
|
| Look again at ex 7.11 and you see that the Count statement does not appear within the Model command. Also take a look at our Short Course videos and handouts on our website. When you have problems running Mplus, send your output to support@statmodel.com along with your license number. |
|
|
|
|
Thank you for replying.
I have sent the imput to the support@statmodel.com, please confirm. |
|
|
| Back to top |
|
|
