Sometimes, you might have a priori information — perhaps, from an external source — on the class membership of some individuals. For example, in a four-class situation, one may know that case 5 belongs to latent class 2 and case 11 to latent class 3. Similarly, you may have a priori information on which class cases do not belong. For example, again in a four-class situation, you may know that case 19 does not belong to latent class 2 and that case 41 does not belong to latent classes 3 or 4. In LatentGOLD, there is an option — called “Known Class” — for indicating to which latent classes cases do not belong. For technical details as to how the likelihood function is modified to incorporate this option see section 2.5 of the LatentGOLD Technical Guide.
Common applications include:
1. Multiple group models (we will explore an example of this below and in Exercise A)
2. using new data to refine old segmentation models while maintaining the segment classifications of the original sample
3. archetypal analysis — define class membership a priori based on extreme response patterns that reflect theoretical “archetypes”
4. partial classification — high cost (or other factors) may preclude all but a small sample of cases from being classified with certainty. These cases can be assigned to their respective classes with 100% certainty, and the remaining would be classified by the LC model in the usual way
5. certain cases may be known to be “type 1 OR type 2″ (e.g., ‘clinically depressed’ or `troubled’). By excluding such cases from being in say class 3 = ‘healthy’, such cases can be pre-assigned to be in class 1 or 2, while additional cases may be freely classified into any class
6. post-hoc refinement of class assignment where modal assignment for certain cases is judged to be implausible based on the desired interpretation of the classes.
” download=””>candy.dat
Multiple group example: In LatentGOLD open the file gss82.sav and specify a full (complete heterogeneity) 2-group LC cluster model that is equivalent to specifying separate 3-class models for the White and Black samples. To accomplish this, follow the following steps:
1. Include the 4 indicators and specify them as Nominal.
2. Specify 6 latent classes.
3. In the ClassPred Tab specify the variable RACE as the known class indicator and select SCAN.
4. Assign checkmarks so that Whites are assigned to classes 1,2 and 3 and Blacks to classes 4, 5, and 6.
5. Estimate the model.
See Exercise A.
Important Note: In practice, these types of multiple group models are not used much anymore, largely replaced by the grouping variable being used as a nominal covariate in the LatentGOLD GUI.
For attendees who want to pursue these type of models further, it is much easier to estimate Multiple Group Models using the LatentGOLD Syntax module. As an example of doing this, from the HELP menu select:
“Syntax Examples > Latent Class Analysis > Multiple Group LCA > Polytomous”, which will open the saved demo model ‘Gss82_multigroup.lgs’ with the same demo file ‘gss82.sav’ used in Exercise A.