EM_algo_c: Parameter estimation via EM Algorithm with Condition Stage
In iotarelr: Iota Inter Coder Reliability for Content Analysis

EM_algo_c

R Documentation

Parameter estimation via EM Algorithm with Condition Stage

Description

Function written in C++ for estimating the parameters of the model via Expectation Maximization (EM Algorithm).

Usage

EM_algo_c(
  obs_pattern_shape,
  obs_pattern_frq,
  obs_internal_count,
  categorical_levels,
  random_starts,
  max_iterations,
  rel_convergence,
  con_step_size,
  con_random_starts,
  con_max_iterations,
  con_rel_convergence,
  fast,
  trace,
  con_trace
)

Arguments

`obs_pattern_shape`	`Matrix` containing the unique patterns found in the data. Ideally this matrix is generated by the function `get_patterns()`.
`obs_pattern_frq`	`Vector` containing the frequencies of the patterns. Ideally it is generated by the the function `get_patterns()`.
`obs_internal_count`	`Matrix` containing the relative frequencies of each category within each pattern. Ideally this matrix is generated by the function `get_patterns()`.
`categorical_levels`	`Vector` containing all possible categories of the content analysis.
`random_starts`	`Integer` for determining how often the algorithm should restart with randomly chosen values for the Assignment Error Matrix and the categorical sizes.
`max_iterations`	`Integer` for determining the maximum number of iterations for each random start.
`rel_convergence`	`Double` for determining the convergence criterion. The algorithm stops if the relative change is smaller than this criterion.
`con_step_size`	`Double` for specifying the size for increasing or decreasing the probabilities during the condition stage of estimation. This value should not be less than 1e-3.
`con_random_starts`	`Integer` for the number of random starts within the condition stage.
`con_max_iterations`	`Integer` for the maximum number of iterations during the condition stage.
`con_rel_convergence`	`Double` for determining the convergence criterion during condition stage. The algorithm stops if the relative change is smaller than this criterion.
`fast`	`Bool` If `TRUE` a fast estimation is applied during the condition stage. This option ignores all parameters beginning with "con_". If `FALSE` the estimation described in Berding and Pargmann (2022) is used. Default is `TRUE`.
`trace`	`TRUE` for printing progress information on the console. `FALSE` if this information should not be printed.
`con_trace`	`TRUE` for printing progress information on the console during estimations in the condition stage. `FALSE` if this information should not be printed.

Value

Function returns a list with the estimated parameter sets for every random start. Every parameter set contains the following components:

`log_likelihood`	Log likelihood of the estimated solution.
`aem`	Estimated Assignment Error Matrix (aem). The rows represent the true categories while the columns stand for the assigned categories. The cells describe the probability that a coding unit of category i is assigned to category j.
`categorial_sizes`	`Vector` of estimated sizes for each category.
`convergence`	If the algorithm converged within the iteration limit `TRUE`. `FALSE` in every other case.
`iteration`	Number of iterations when the algorithm was terminated.

References

Berding, Florian, and Pargmann, Julia (2022).Iota Reliability Concept of the Second Generation.Measures for Content Analysis Done by Humans or Artificial Intelligences. Berlin: Logos. https://doi.org/10.30819/5581

iotarelr documentation built on May 29, 2024, 7:33 a.m.