The maximum likelihood threshold of a path diagram (Q2414096)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | The maximum likelihood threshold of a path diagram |
scientific article |
Statements
The maximum likelihood threshold of a path diagram (English)
0 references
10 May 2019
0 references
For a random vector $X=(X_1, \ldots , X_p)$ a linear structural equation model incoporates linear constraints of the form: \[ X_i = \lambda_{0i} + \sum_{j \ne i} \lambda_{ij}X_j + \varepsilon_i, \quad i=1, \dots, p, \] where $\varepsilon$ is a random error vector, assumed to be normal with zero mean and variance-covariance matrix $\Omega=(\omega_{ij})$, and the $\lambda$s are real parameters. The structure of the model is summarized in a mixed graph $(V,D,B)$ where: $V=\{1, \dots, p\}$ represents the set of variables; $D \subset V \times V$ is a directed graph that captures the structure of the $\lambda$ coefficients ($i \rightarrow j \notin D$ if and only if $\lambda_{ij} = 0$); and $B$ is a set containing $2$-element subsets (bidirected edges) of $V$ describing the correlation structure of $\varepsilon$ ($\omega_{ij} = 0$ when $i \ne j$ and $i \leftrightarrow j \notin B$). \par In this paper, the authors compute the maximum likelihood threshold (mlt) of such models. The mlt is the minimum sample size $N$ such that the likelihood function is bounded a.s. for any sample size $n \ge N$. The resulting formula has a simple expression, since it is enough to take all the connected components $C_1, \dots, C_l$ of the bidirected part of the mixed graph, and to compute, for each $C_j$, the cardinality of the union of $C_j$ with the parents of its elements.
0 references
covariance matrix
0 references
graphical model
0 references
maximum likelihood
0 references
normal distribution
0 references
path diagram
0 references
structural equation model
0 references