good-practice

Base fitting functions in R will seek variables in the global environment (or more generally in the environment where a call to <code>`~`</code> was made, defining the model formula) if they are not in the <code>data</code>. This easily leads to errors (see example in the discussion of <code><a rd-options="" href="/link/update.HLfit?package=spaMM&version=2.0.0" data-mini-rdoc="spaMM::update.HLfit">update.HLfit</a></code>). Indeed Chambers (2008, p.221), after describing how the environment is defined, comments that &#8220;Where clear and trustworthy software is a priority, I would personally avoid such tricks. Ideally, all the variables in the model frame should come from an explicit, verifiable data source...&#8221;. Hence, the main fitting functions in <code>spaMM</code> depart from the sloppy practice. They strip the formula environment from any variable, and seek all variables from the <code>formula</code> in the data frame given by their <code>data</code> argument. <b>One never needs to specify the <code>data</code> in the <code>formula</code>.</b> The variables defining the <code>prior.weights</code> should also be in the <code>data</code>. Variables used in other arguments such as <code>ranFix</code> are looked up neither in the data nor in the formula environment, but in the calling environment as usual.

Inference in mixed-effect models, including generalized linear mixed models with spatial
correlations and models with non-Gaussian random effects (e.g., Beta Binomial,
or negative-binomial mixed models). Variation in residual variance is handled and can be modelled
as a linear model. The algorithms are currently various Laplace approximations
methods for likelihood or restricted likelihood, in particular h-likelihood and penalized-likelihood
methods.

Franois Rousset

spaMM

Mixed-Effect Models, Particularly Spatial Models

Fran<c3><a7>ois Rousset

Jean-Baptiste Ferdy

Alexandre Courtiol

GSL authors 

good-practice function

Base fitting functions in R will seek variables in the global environment (or more generally in the environment where a call to <code>`~`</code> was made, defining the model formula) if they are not in the <code>data</code>. This easily leads to errors (see example in the discussion of <code><a rd-options='' href='update.HLfit'>update.HLfit</a></code>). Indeed Chambers (2008, p.221), after describing how the environment is defined, comments that &#8220;Where clear and trustworthy software is a priority, I would personally avoid such tricks. Ideally, all the variables in the model frame should come from an explicit, verifiable data source...&#8221;. Hence, the main fitting functions in <code>spaMM</code> depart from the sloppy practice. They strip the formula environment from any variable, and seek all variables from the <code>formula</code> in the data frame given by their <code>data</code> argument. <b>One never needs to specify the <code>data</code> in the <code>formula</code>.</b> The variables defining the <code>prior.weights</code> should also be in the <code>data</code>. Variables used in other arguments such as <code>ranFix</code> are looked up neither in the data nor in the formula environment, but in the calling environment as usual.

good-practice: Clear and trustworthy formulas

Description

Arguments

References