hopit.control

an epsilon parameter ("a very small number") used to calculate the Hessian from the gradient function.

grad.eps

the maximum number of iterations.
See <code><a rd-options="" href="/link/optim?package=hopit&version=0.9.0" data-mini-rdoc="hopit::optim">optim</a></code> and <code><a rd-options="" href="/link/nlm?package=hopit&version=0.9.0" data-mini-rdoc="hopit::nlm">nlm</a></code> for details.

bgfs.maxit, cg.maxit, nlm.maxit

the relative convergence tolerances for the BFGS and the CG methods.
See <code><a rd-options="" href="/link/optim?package=hopit&version=0.9.0" data-mini-rdoc="hopit::optim">optim</a></code> for details.

bgfs.reltol, cg.reltol

a tolerance at which the scaled gradient is
considered close enough to zero and
a minimum allowable relative step length for the nlm method. See <code><a rd-options="" href="/link/nlm?package=hopit&version=0.9.0" data-mini-rdoc="hopit::nlm">nlm</a></code>.

nlm.gradtol, nlm.steptol

"CG", "BFGS", or both. If both, the CG is run first, followed by the BFGS. See <code><a rd-options="" href="/link/optim?package=hopit&version=0.9.0" data-mini-rdoc="hopit::optim">optim</a></code>.

fit.methods

a logical; if FALSE (default) the <code>nlm</code> optimization method
is omitted and only the BFGS and/or the CG methods are run.

nlm.fit

a logical for whether to trace the process of model fitting.

trace

a type of transformation applied to
the all of the latent's or all of the threshold's numeric variables. Possible values:<ul>
<li>"none" - no transformation</li>
<li>"min" - subtract the minimum from a variable</li>
<li>"scale_01" - transform the variable to fit the range from 0 to 1</li>
<li>"standardize" or "standardise" - subtract the mean from a variable then divide it by it's standard deviation</li>
<li>"standardize_trunc" or "standardise_trunc" - subtract the minimum from a variable then divide it by it's standard deviation</li>
</ul>

transform.latent, transform.thresh

An auxiliary function for controlling the fitting of a <code>hopit</code> model.
Use this function to set the control
parameters of the <code><a rd-options="" href="/link/hopit?package=hopit&version=0.9.0" data-mini-rdoc="hopit::hopit">hopit</a></code> and other related functions.

Self-reported health, happiness, attitudes, and other statuses or
perceptions are often the subject of biases that may come from different
sources. For example, the evaluation of own health may depend on previous
medical diagnoses, functional status, and symptoms and signs of illness, as
well as life-style behaviors including contextual social, gender, age-specific,
linguistic and other cultural factors (Jylha 2009
<doi:10.1016/j.socscimed.2009.05.013>; Oksuzyan et al. 2019
<doi:10.1016/j.socscimed.2019.03.002>). This package offers versatile functions
for analyzing different self-reported ordinal variables and helping to estimate
their biases.
Specifically, the package provides the function to fit a generalized ordered
probit model that regresses original self-reported status measures on two sets
of independent variables (King et al. 2004 <doi:10.1017/S0003055403000881>;
Jurges 2007 <doi:10.1002/hec.1134>; Oksuzyan et al. 2019
<doi:10.1016/j.socscimed.2019.03.002>). In contrast to standard ordered probit
models, generalized ordered probit models relax the assumption that individuals
use a common scale when rating their own statuses, and thus allow for
distinguishing between the status (e.g., health) and reporting differences
based on other individual characteristics. In other words, the model accounts
for heterogeneity in reporting behaviors. The first set of variables
(e.g., health variables) included in the regression are individual statuses and
characteristics that are directly related to the self-reported variable.
In case of self-reported health, these could be chronic conditions, mobility
level, difficulties with daily activities, performance on grip strength tests,
anthropometric measures, and lifestyle behaviors. The second set of independent
variables (threshold variables) is used to model cut-points between adjacent
self-reported response categories as functions of individual characteristics,
such as gender, age group, education, and country
(Oksuzyan et al. 2019 <doi:10.1016/j.socscimed.2019.03.002>).
The model helps adjust for these specific socio-demographic and cultural
differences in how the continuous latent health is projected onto the ordinal
self-rated measure.
The fitted model can be used to calculate an individual latent status variable
that serves as a proxy of the true status. In case of self-reported health, the
predicted latent health variable can be standardized to a health index, which
varies from 0 representing the (model-based) worst health state to 1
representing the (model-based) best health in the sample.
The standardized latent coefficients (disability weights for the case of
self-rated health) provide information about the individual impact of the
specific latent (e.g., health) variables on the latent (e.g., health)
construct. For example, they indicate the extent to which the latent health
index is reduced by the presence of Parkinson<e2><80><99>s disease, poor mobility, and
other specific health measures (Jurges 2007 <doi:10.1002/hec.1134>; Oksuzyan
et al. 2019 <doi:10.1016/j.socscimed.2019.03.002>).
The latent index can in turn be used to reclassify the categorical status
measure that has been adjusted for inter-individual differences in reporting
behavior. Two methods for doing so are available, one which uses model
estimated cut-points, and a second which reclassifies responses according to
the percentiles of the original categorical response distribution
(Jurges 2007 <doi:10.1002/hec.1134>;
Oksuzyan et al. 2019 <doi:10.1016/j.socscimed.2019.03.002>).

Maciej J. Danko

hopit

Hierarchical Ordered Probit Models with Application to Reporting
Heterogeneity

hopit.control function

the maximum number of iterations.
See <code><a rd-options='' href='optim'>optim</a></code> and <code><a rd-options='' href='nlm'>nlm</a></code> for details.

the relative convergence tolerances for the BFGS and the CG methods.
See <code><a rd-options='' href='optim'>optim</a></code> for details.

a tolerance at which the scaled gradient is
considered close enough to zero and
a minimum allowable relative step length for the nlm method. See <code><a rd-options='' href='nlm'>nlm</a></code>.

"CG", "BFGS", or both. If both, the CG is run first, followed by the BFGS. See <code><a rd-options='' href='optim'>optim</a></code>.

An auxiliary function for controlling the fitting of a <code>hopit</code> model.
Use this function to set the control
parameters of the <code><a rd-options='' href='hopit'>hopit</a></code> and other related functions.

hopit.control: Auxiliary for controlling the fitting of a `hopit` model

Description

Usage

Arguments

See Also