1: \begin{abstract}

2: Hierarchical parametric models consisting of observable and latent variables

3: are widely used for unsupervised learning tasks.

4: For example, a mixture model is a representative hierarchical model for clustering.

5: From the statistical point of view,

6: the models can be regular or singular due to the distribution of data.

7: In the regular case, the models have the identifiability;

8: there is one-to-one relation

9: between a probability density function for the model expression and the parameter.

10: The Fisher information matrix is positive definite, and

11: the estimation accuracy of both observable and latent variables has been studied.

12: In the singular case,

13: on the other hand, the models are not identifiable and the Fisher matrix is not positive definite.

14: Conventional statistical analysis based on the inverse Fisher matrix is not applicable.

15: Recently, an algebraic geometrical analysis has been developed

16: and is used to elucidate the Bayes estimation of observable variables.

17: The present paper applies this analysis to latent-variable estimation

18: and determines its theoretical performance.

19: Our results clarify behavior of the convergence of the posterior distribution.

20: It is found that the posterior of the observable-variable estimation

21: can be different from the one in the latent-variable estimation.

22: Because of the difference, the Markov chain Monte Carlo method

23: based on the parameter and the latent variable

24: cannot construct the desired posterior distribution.

25: %\end{abstract}