add_wording_effects {latentFactoR}

Adds Wording Effects to simulate_factors Data

Description

Adds wording effects to simulated data from simulate_factors. See examples to get started

Usage

add_wording_effects(
  lf_object,
  method = c("acquiescence", "difficulty", "random_careless", "straight_line", "mixed"),
  proportion_negative = 0.5,
  proportion_negative_range = NULL,
  proportion_biased_cases = 0.1,
  proportion_biased_variables = 1,
  proportion_biased_variables_range = NULL,
  proportion_biased_person = 1,
  proportion_biased_person_range = NULL
)

Arguments

lf_object

Data object from simulate_factors. Data must be categorical. If data are not categorical, then there function with throw an error

method

Character (length = 1). Method to generate wording effect to add to the data. Description of methods:

• "acquiescence" —Generates new data with flipped dominant loadings (based on proportion_negative) and ensures a bias such that variables have a restricted range of responding (e.g., only 4s and 5s on a 5-point Likert scale)

• "difficulty" — Generates new data with flipped dominant loadings (based on proportion_negative) and uses this data as the data without wording effects. Then, the signs of the dominant loadings are obtained and the dominant loadings are made to be absolute. Finally, the skews are multiplied by the signs of the original dominant loadings when generating the data with the wording effects

• "random_careless" — Number of cases up to proportion_biased_cases are sampled and replaced by values from a random uniform distribution ranging between the lowest and highest response category for each variable. These values then replace the values in the original data

• "straight_line" — Coming soon...

proportion_negative

Numeric (length = 1 or factors). Proportion of variables that should have negative (or flipped) dominant loadings across all or each factor. Accepts number of variables as well. The first variables on each factor, up to the corresponding proportion, will be flipped. Set to 0 to not have any loadings flipped. Defaults to 0.50

proportion_negative_range

Numeric (length = 2). Range of proportion of variables that are randomly selected from a uniform distribution. Accepts number of number of variables as well. Defaults to NULL

proportion_biased_cases

Numeric (length = 1). Proportion of cases that should be biased with wording effects. Also accepts number of cases to be biased. The first n number of cases, up to the corresponding proportion, will be biased. Defaults to 0.10 or 10 percent of cases.

proportion_biased_variables

Numeric (length = 1 or factors). Proportion of variables that should be biased with wording effects. For method = "difficulty", proportion of biased variables will only count for the negative variables. For method = "acquiescence", proportion of biased variables will only count for variables below the mid-point of the variable_categories. Defaults to 1 or all possible variables

proportion_biased_variables_range

Numeric (length = 2). Range of proportion of variables that should be biased with wording effects. Values are drawn randomly from a uniform distribution. Defaults to NULL

proportion_biased_person

Numeric (length = 1 or proportion_biased_cases x sample_size). Person-specific parameter of how many much bias the proportion_biased_cases will have over the possible biased variables. This parameter interacts with proportion_biased_variables. Parameter specifies the proportion of variables that should have bias per person. If one value is provided, then all biased cases will have the same proportion of variables biased. Individual values are possible by providing values for each biased case (round(nrow(lf_object$data) * proportion_biased_cases)). Setting individual values for each biased case is not recommended (use proportion_biased_person_range instead). Defaults to 1 or all possible biased variables for all biased cases

proportion_biased_person_range

Numeric (length = 2). Range to randomly draw bias from a uniform distribution. Allows for random person-specific bias to be obtained. Defaults to NULL

Value

Returns a list containing:

data

Biased data simulated data from the specified factor model

unbiased_data

The corresponding unbiased data prior to replacing values to generate the (biased) data

biased_sample_size

The number of cases that have biased data

adjusted_results

Bias-adjusted lf_object input into function

original_results

Original lf_object input into function

Author(s)

Alexander P. Christensen <[e-mail]>, Luis Eduardo Garrido <[e-mail]>

References

Garcia-Pardina, A., Abad, F. J., Christensen, A. P., Golino, H., & Garrido, L. E. (2022). Dimensionality assessment in the presence of wording effects: A network psychometric and factorial approach. PsyArXiv.

Garrido, L. E., Golino, H., Christensen, A. P., Martinez-Molina, A., Arias, V. B., Guerra-Pena, K., ... & Abad, F. J. (2022). A systematic evaluation of wording effects modeling under the exploratory structural equation modeling framework. PsyArXiv.

Examples

# Generate factor data
two_factor <- simulate_factors(
  factors = 2, # factors = 2
  variables = 6, # variables per factor = 6
  loadings = 0.55, # loadings between = 0.45 to 0.65
  cross_loadings = 0.05, # cross-loadings N(0, 0.05)
  correlations = 0.30, # correlation between factors = 0.30
  sample_size = 1000, # number of cases = 1000
  variable_categories = 5 # 5-point Likert scale
)

# Add wording effects using acquiescence method
two_factor_acquiescence <- add_wording_effects(
  lf_object = two_factor,
  proportion_negative = 0.50,
  proportion_biased_cases = 0.10,
  method = "acquiescence"
)

# Add wording effects using difficulty method
two_factor_difficulty <- add_wording_effects(
  lf_object = two_factor,
  proportion_negative = 0.50,
  proportion_biased_cases = 0.10,
  method = "difficulty"
)

# Add wording effects using random careless method
two_factor_random_careless <- add_wording_effects(
  lf_object = two_factor,
  proportion_negative = 0.50,
  proportion_biased_cases = 0.10,
  method = "random_careless"
)

# Add wording effects using straight line method
two_factor_random_careless <- add_wording_effects(
  lf_object = two_factor,
  proportion_negative = 0.50,
  proportion_biased_cases = 0.10,
  method = "straight_line"
)

# Add wording effects using mixed method
two_factor_mixed <- add_wording_effects(
  lf_object = two_factor,
  proportion_negative = 0.50,
  proportion_biased_cases = 0.10,
  method = "mixed"
)

# Add wording effects using acquiescence and straight line method
two_factor_multiple <- add_wording_effects(
  lf_object = two_factor,
  proportion_negative = 0.50,
  proportion_biased_cases = 0.10,
  method = c("acquiescence", "straight_line")
)