Package 'SurrogateRsq'

Red wine quality dataset of the Portuguese "Vinho Verde" wine

Description

A red wine tasting preference data used in the study of Cortez, Cerdeira, Almeida, Matos, and Reis 2009. This red wine contains 1599 samples and 12 variables including the tasting preference score of red wine and its physicochemical characteristics.

Usage

data(RedWine)

Format

A data frame with 1599 rows, quality score, and 11 variables of physicochemical properties of wines.

• quality Tasting preference is a rating score provided by a minimum of three sensory with ordinal values from 0 (very bad) to 10 (excellent). The final sensory score is the median of these evaluations.

• fixed.acidity The fixed acidity is the physicochemical property in unit (g(tartaric acid)/dm^3).

• volatile.acidity The volatile acidity is in unit g(acetic acid)/dm^3.

• citric.acid The citric acidity is in unit g/dm^3.

• residual.sugar The residual sugar is in unit g/dm^3.

• chlorides The chlorides is in unit g(sodium chloride)/dm^3.

• free.sulfur.dioxide The free sulfur dioxide is in unit mg/dm^3.

• total.sulfur.dioxide The total sulfur dioxide is in unit mg/dm^3.

• density The density is in unit g/cm^3.

• pH The wine's pH value.

• sulphates The sulphates is in unit g(potassium sulphates)/dm^3.

• alcohol The alcohol is in unit \

References

Cortez, P., Cerdeira, A., Almeida, F., Matos, T., and Reis, J. (2009), “Modeling wine preferences by data mining from physicochemical properties,” Decision Support Systems, 47, 547–553. doi:10.1016/j.dss.2009.05.016

Examples

head(RedWine)

---

A function to calculate the surrogate R-squared measure.

Description

It can provide the surrogate R-squared for a user specified model. This function will generate an S3 object of surrogate R-squared measure that will be called from other functions of this package. The generic S3 function print is also developed to present the surrogate R-squared measure.

Usage

surr_rsq(model, full_model, avg.num = 30, asym = FALSE, newdata = NULL, ...)

Arguments

model	A reduced model that needs to be investigated. The reported surrogate R-squared is for this reduced model.
full_model	A full model that contains all of the predictors in the data set. This model object should also contain the dataset for fitting the full model and the reduced model in the first argument.
avg.num	The number of replication for the averaging of surrogate R-squared.
asym	A logical argument whether use the asymptotic version of our surrogate R-squared. More details are in the paper Liu et al. (2023).
...	Additional optional arguments.

Value

An object of class "surr_rsq" is a list containing the following components:

surr_rsq	the surrogate R-squared value;
reduced_model	the reduced model under investigation. It should be a subset of the full model;
full_model	the full model used for generating the surrogate response. It should have passed initial variable screening and model diagnostics (see Paper for reference);
data	the dataset contains the response variable and all the predictors.

References

Zhu, X., Liu, D., Lin, Z., Greenwell, B. (2022). SurrogateRsq: an R package for categorical data goodness-of-fit analysis using the surrogate R-squared

Liu, D., Zhu, X., Greenwell, B., & Lin, Z. (2023). A new goodness‐of‐fit measure for probit models: Surrogate R2. British Journal of Mathematical and Statistical Psychology, 76(1), 192-210.

Examples

data("RedWine")

full_formula <- as.formula(quality ~ fixed.acidity + volatile.acidity +
citric.acid+ residual.sugar + chlorides + free.sulfur.dioxide +
total.sulfur.dioxide + density + pH + sulphates + alcohol)

full_mod <- polr(formula = full_formula,
data=RedWine, method  = "probit")

select_model <- update(full_mod, formula. = ". ~ . - fixed.acidity -
citric.acid - residual.sugar - density")
surr_obj_sele_mod <- surr_rsq(model = select_model, full_model = full_mod,
                               data = RedWine, avg.num = 30)
print(surr_obj_sele_mod$surr_rsq, digits = 3)

---

A function to calculate the interval estimate of the surrogate R-squared measure

Description

This function generates the interval measure of surrogate R-squared by bootstrap.

Usage

surr_rsq_ci(
  object,
  alpha = 0.05,
  B = 2000,
  asym = FALSE,
  parallel = FALSE,
  ...
)

Arguments

object	A object of class "surr_rsq" that is generated by the function "surr_rsq". It contains the following components: surr_rsq, reduced_model, full_model, and data.
alpha	The significance level alpha. The confidence level is 1-alpha.
B	The number of bootstrap replications.
asym	A logical argument whether use the asymptotic version of our surrogate R-squared. More details are in the paper of Liu et al. (2023).
parallel	logical argument whether conduct parallel for bootstrapping surrogate R-squared to construct the interval estimate. The clusters need to be registered through registerDoParallel(cl) beforehand.
...	Additional optional arguments.

Value

An list that contains the CI_lower, CI_upper.

Examples

data("RedWine")

full_formula <- as.formula(quality ~ fixed.acidity + volatile.acidity + citric.acid
+ residual.sugar + chlorides + free.sulfur.dioxide +
total.sulfur.dioxide + density + pH + sulphates + alcohol)

fullmodel <- polr(formula = full_formula,data=RedWine, method  = "probit")

select_model <- update(fullmodel, formula. = ". ~ . - fixed.acidity -
citric.acid - residual.sugar - density")

surr_rsq_select <- surr_rsq(select_model, fullmodel, data = RedWine, avg.num = 30)

# surr_rsq_ci(surr_rsq_select, alpha = 0.05, B = 2000, parallel = FALSE) # Not run, it takes time.

# surr_rsq_ci(surr_rsq_select, alpha = 0.05, B = 2000, parallel = TRUE) # Not run, it takes time.

---

The contribution of each variable in the final model

Description

This function calculates reduction of the surrogate R-squared goodness-of-fit of each variable to measure their relative explanatory power. This function creates a table containing the reductions of surrogate R-squared by removing each one of variables in the model.

Usage

surr_rsq_rank(object, avg.num = 30, var.set = NA, ...)

Arguments

object	A object of class "surr_rsq" that is generated by the function "surr_rsq". It contains the following components: surr_rsq, reduced_model, full_model, and data.
avg.num	The number of replication for the averaging of surrogate R-square.
var.set	A list that contains a few sets. Each component of these sets represents the variables that you want to examine for the contribution of goodness of fit. Then, for one component of this list, a model will fit by removing the specified variables.
...	Additional optional arguments.

Value

The default return is a list that contains the contribution of Surrogate R-squared for each variable in the full_model. If the var.set is specified, the return is a list of the contribution of the groups of variables in the var.set.

Examples

data("WhiteWine")

sele_formula <- as.formula(quality ~ fixed.acidity + volatile.acidity +
                          residual.sugar +  + free.sulfur.dioxide +
                          pH + sulphates + alcohol)

sele_mod <- polr(formula = sele_formula,
              data = WhiteWine,
              method = "probit")

sur1 <- surr_rsq(model = sele_mod,
              full_model = sele_mod,
              avg.num = 100)

rank_tab_sur1 <- surr_rsq_rank(object  = sur1,
                               avg.num = 30)
print(rank_tab_sur1)

---

White wine quality dataset of the Portuguese "Vinho Verde" wine

Description

A white wine tasting preference data used in the study of Cortez, Cerdeira, Almeida, Matos, and Reis 2009. This white wine contains 4898 white vinho verde wine samples and 12 variables including the tasting preference score of white wine and its physicochemical characteristics.

Usage

data(WhiteWine)

Format

A data frame with 4898 rows, quality score, and 11 variables of physicochemical properties of wines.

• quality Tasting preference is a rating score provided by a minimum of three sensory with ordinal values from 0 (very bad) to 10 (excellent). The final sensory score is the median of these evaluations.

• fixed.acidity The fixed acidity is the physicochemical property in unit (g(tartaric acid)/dm^3).

• volatile.acidity The volatile acidity is in unit g(acetic acid)/dm^3.

• citric.acid The citric acidity is in unit g/dm^3.

• residual.sugar The residual sugar is in unit g/dm^3.

• chlorides The chlorides is in unit g(sodium chloride)/dm^3.

• free.sulfur.dioxide The free sulfur dioxide is in unit mg/dm^3.

• total.sulfur.dioxide The total sulfur dioxide is in unit mg/dm^3.

• density The density is in unit g/cm^3.

• pH The wine's pH value.

• sulphates The sulphates is in unit g(potassium sulphates)/dm^3.

• alcohol The alcohol is in unit \

References

Cortez, P., Cerdeira, A., Almeida, F., Matos, T., and Reis, J. (2009), “Modeling wine preferences by data mining from physicochemical properties,” Decision Support Systems, 47, 547–553. doi:10.1016/j.dss.2009.05.016

Examples

head(WhiteWine)

---