poisson([Model] [, print:F or silent:T, incr:T, offsets:vec, pvals:T,\ maxiter:m, epsilon:eps, coefs:F]), vec a REAL vector, m an integer > 0, eps REAL > 0 |

poisson(Model) computes a log linear regression fit of the model specified in the CHARACTER variable Model. If y is the response variable in the model it must be a REAL vector with y[i] >= 0. Estimation is by maximum likelihood on the assumption that y[i] is Poisson. If any y[i] is not an integer a warning message is printed. See topic 'models' for information on specifying Model. poisson(Model,...) is equivalent to glmfit(Model,dist:"poisson", link:"log",...). poisson() sets the side effect variables RESIDUALS, WTDRESIDUALS, SS, DF, HII, DEPVNAME, TERMNAMES, and STRMODEL. The elements of WTDRESIDUALS are the final weighted residuals in the iteratively reweighted least squares fit to log(response). See topic 'glm'. Without keyword phrase 'inc:T' (see below), TERMNAMES has value vector("","", ...,"Overall model","ERROR1"), DF has value vector(0,0, ...,ModelDF,ErrorDF) and SS has value vector(0,0,...,ModelDeviance, ErrorDeviance). If, say, Model is "y=x1+x2", an iterative algorithm fits log(y) as a linear function of x1 and x2. A two line Analysis of Deviance table is printed, with line 1 the diffence between the deviance from a model with all coefficients 0 and the deviance of the estimated model, and line 2, labeled "ERROR", the deviance of the estimated model. Under appropriate assumptions, the latter can be used to test the goodness of fit of the model. poisson(Model,inc:T) computes the full Poisson model and all partial models -- only a constant term, the constant and the first term, and so on. It prints an Analysis of Deviance table, with one line for each term, plus the deviance of the complete model labeled as "ERROR" . Each term's deviance is the reduction in deviance associated with that term. If you omit Model (poisson()), the model from the most recent GLM command such as poisson() or anova(), or the model in CHARACTER variable STRMODEL is assumed. Computations are carried out using iteratively reweighted least squares with starting values derived from an unweighted least squares fit of log(y + .25). Other keyword phrases Keyword phrase Default Meaning maxiter:m 50 Positive integer m is the maximum number of iterations that will be allowed in fitting epsilon:eps 1e-6 Small positive REAL specifying relative error in objective function (2*log likelihood) required to end iteration offsets:OffVec none Causes model to be fit to log(p) to be 1*Offvec + Model, where OffVec is a REAL vector the same length as response y. Note OffVec is in log units. See topic 'glm_keys' for information on keyword phrases print:F, silent:T, coefs:F and pvals:T. The default value for pvals can be changed by setoptions(pvals:T). See topic setoptions(), subtopic 'options:"pvals"'. Examples of the use of 'offsets'. Cmd> poisson("y=x", offsets:3*x, inc:T, pvals:T) The P value associated with x can be used to test the hypothesis H0: beta1 = 3 in the model log(E[y]) = beta0 + beta1*x. Cmd> poisson("y=1", offsets:rep(log(10), length(y), inc:T, pvals:T) The P value associated with the CONSTANT term can be used to test H0: E[y] = 10, assuming y contains a random sample from a Poisson distribution.

Gary Oehlert 2003-01-15