NCLVAL — Vector of length NCLVAR containing, in its i‑th element, the number of levels or categories of the i‑th classification variable. (Input)

TABLE — Vector of length NCLVAL(1) * NCLVAL(2) * … * NCLVAL(NCLVAR) containing the entries in the cells of the table to be fit. (Input)
See Comment 3 for comments on the ordering of the elements of TABLE.

NVEF — Vector of length NEF containing the number of classification variables associated with each effect. (Input)

INDEF — Vector of length NVEF(1) + … + NVEF(NEF) containing, in consecutive positions, the indices of the variables that are included in each effect. (Input)
The entries in INDEF are sequenced so that the first NVEF(1) elements contain the indices of the variables in effect 1, the next NVEF(2) elements of INDEF contain the indices of the variables in effect 2, etc. See Comment 4 for an example.

FIT — Vector of length NCLVAL(1) * NCLVAL(2) * … * NCLVAL(NCLVAR) containing the model estimates of the cell frequencies. (Input/Output)
On input, FIT contains the initial estimates of the cell counts. Structural zeros in the model are specified by setting the corresponding element of FIT to 0.0. All other elements of FIT may be set to 1.0 if no other estimate of the expected cell counts is available. On output, FIT contains the fitted table. See Comment 3 for the ordering of the elements of FIT. If an element of FIT is positive but the corresponding element in TABLE is zero, then the element is called a sampling zero. Sampling zeros may effect the number of parameters that can be estimated, but they will not effect the degrees of freedom in chi‑squared tests. See the Description section.

NCOEF — Number of regression coefficients in the model. (Output)

COEF — NCOEF by 4 matrix containing the estimated coefficients and associated statistics. (Output)
Dummy variables used in fitting the log‑linear model are generated using the IDUMMY = 3 option of routine GRGLM (see Chapter 2, “Regression”). For this option, the k‑th dummy variable for classification variable I is the (0, 1) indicator variable for the k‑th level of the classification variable minus the (0, 1) indicator variable for the NCLVAL(I)-th level of the classification variable.

COV — NCOEF by NCOEF covariance matrix for the estimated parameters. (Output)

RESID — NCLVAL(1) * NCLVAL(2) * … * NCLVAL(NCLVAR) by 4 matrix containing residual statistics for each cell in the table. (Output)

STAT — Vector of length 4 containing output statistics for the model. (Output)

NCLVAR — Number of classification variables. (Input)
A variable specifying a margin in the table is a classification variable. The first classification variable is named A, the second classification variable is named B, etc.
Default: NCLVAR = size (NCLVAL,1).

NEF — Number of effects in the model. (Input)
A marginal table is implied by each effect in the model. Lower‑order effects should not be included since their inclusion is automatic in the hierarchical models fit here (e.g., do not include effects A or B if effect AB is in the model).
Default: NEF = size (NVEF,1).

EPS — Convergence criterion. (Input)
Convergence is assumed when the maximum deviation between an observed and a fitted marginal total is less than EPS. EPS = 0.10 is a typical value.
Default: EPS = 0.10.

MAXIT — Maximum number of iterations. (Input)
MAXIT = 15 is a typical value.
Default: MAXIT = 30

TOL — Tolerance used in determining linear dependence in COV. (Input)
TOL = 100.0 AMACH(4) is a common choice. See the documentation for routine AMACH in Reference Material.
Default: TOL = 1.19e‑5 for single precision and 2.d–14 for double precision.

IPRINT — Printing option. (Input)
Default: IPRINT = 0.

LDCOEF — Leading dimension of COEF exactly as specified in the dimension statement in the calling program. (Input)
Default: LDCOEF = size (COEF,1).

LDCOV — Leading dimension of COV exactly as specified in the dimension statement in the calling program. (Input)
Default: LDCOV = size (COV,1).

LDRESI — Leading dimension of RESID exactly as specified in the dimension statement in the calling program. (Input)
Default: LDRESI = size (RESID,1).

1. Workspace may be explicitly provided, if desired, by use of C2LLN/DC2LLN. The reference is:

2. Informational errors

3. The cells of the vectors TABLE and ZERO are sequenced so that the first variable cycles from 1 to NCLVAL(1) the slowest, the second variable cycles from 1 to NCLVAL(2) the next slowest, etc., up to the NCLVAR‑th variable, which cycles from 1 to NCLVAL(NCLVAR) the fastest.

Example: For NCLVAR = 3, NCLVAL(1) = 2, NCLVAL(2) = 3, and NCLVAL(3) = 2, the cells of table X(I, J, K) are entered into TABLE(1) through TABLE(12) in the following order: X(1, 1, 1), X(1, 1, 2), X(1, 2, 1), X(1, 2, 2), X(1, 3, 1), X(1, 3, 2), X(2, 1, 1), X(2, 1, 2), X(2, 2, 1), X(2, 2, 2), X(2, 3, 1), X(2, 3, 2). The elements of FIT are similarly sequenced.

4. INDEF is used to describe the marginal tables to be fit. For example, if NCLVAR = 3 and the first effect is to fit the marginal table for variables 1 and 3 and the second effect is to fit the marginal table for variable 2, then: NEF = 2, NVEF(1) = 2, and NVEF(2) = 1. Since the sum of the NVEF(I) is 3, then INDEF is a vector of length 3 with values: INDEF(1) = 1, INDEF(2) = 3, and INDEF(3) = 2.