Package 'rcompanion'

Description

Functions and datasets to support Summary and Analysis of Extension Program Evaluation in R and An R Companion for the Handbook of Biological Statistics.

Useful functions

There are several functions that provide summary statistics for grouped data. These function titles tend to start with "groupwise". They provide means, medians, geometric means, and Huber M-estimators for groups, along with confidence intervals by traditional methods and bootstrap.

Functions to produce effect size statistics, some with bootstrapped confidence intervals, include those for Cramer's V, Cohen's g and odds ratio for paired tables, Cohen's h, Cohen's w, Vargha and Delaney's A, Cliff's delta, r for one-sample, two-sample, and paired Wilcoxon and Mann-Whitney tests, epsilon-squared, and Freeman's theta.

The accuracy function reports statistics for models including minimum maximum accuracy, MAPE, RMSE, Efron's pseudo r-squared, and coefficient of variation.

The functions nagelkerke and efronRSquared provide pseudo R-squared values for a variety of model types, as well as a likelihood ratio test for the model as a whole.

There are also functions that are useful for comparing models. compareLM, compareGLM, and pairwiseModelAnova. These use goodness-of-fit measures like AIC, BIC, and BICc, or likelihood ratio tests.

Functions for nominal data include post-hoc tests for Cochran-Mantel-Haenszel test (groupwiseCMH), for McNemar-Bowker test (pairwiseMcnemar), and for tests of association like Chi-square, Fisher exact, and G-test (pairwiseNominalIndependence).

There are a few useful plotting functions, including plotNormalHistogram that plots a histogram of values and overlays a normal curve, and plotPredy which plots of line for predicted values for a bivariate model. Other plotting functions include producing density plots.

A function close to my heart is cateNelson, which performs Cate-Nelson analysis for bivariate data.

Vignettes and examples

The functions in this package are used in "Extension Education Program Evaluation in R" which is available at https://rcompanion.org/handbook/ and "An R Companion for the Handbook of Biological Statistics" which is available at https://rcompanion.org/rcompanion/.

The documentation for each function includes an example as well.

Version notes

Version 2.0 is not entirely back-compatible as several functions have been removed. These include some of the pairwise methods that can be replaced with better methods. Also, some functions have been removed or modified in order to import fewer packages.

Removed packages are indicated with 'Defunct' in their titles.

Description

Produces a table of fit statistics for multiple models.

Usage

accuracy(fits, plotit = FALSE, digits = 3, ...)

Arguments

Details

Produces a table of fit statistics for multiple models: minimum maximum accuracy, mean absolute percentage error, median absolute error, root mean square error, normalized root mean square error, Efron's pseudo r-squared, and coefficient of variation.

For minimum maximum accuracy, larger indicates a better fit, and a perfect fit is equal to 1.

For mean absolute error (MAE), smaller indicates a better fit, and a perfect fit is equal to 0. It has the same units as the dependent variable. Note that here, MAE is simply the mean of the absolute values of the differences of predicted values and the observed values (MAE = mean(abs(predy - actual))). There are other definitions of MAE and similar-sounding terms.

Median absolute error (MedAE) is similar, except employing the median rather than the mean.

For mean absolute percent error (MAPE), smaller indicates a better fit, and a perfect fit is equal to 0. The result is reported as a fraction. That is, a result of 0.1 is equal to 10 percent.

Root mean square error (RMSE) has the same units as the predicted values.

Normalized root mean square error (NRMSE) is RMSE divided by the mean or the median of the values of the dependent variable.

Efron's pseudo r-squared is calculated as 1 minus the residual sum of squares divided by the total sum of squares. For linear models (lm model objects), Efron's pseudo r-squared will be equal to r-squared. For other models, it should not be interpreted as r-squared, but can still be useful as a relative measure.

CV.prcnt is the coefficient of variation for the model. Here it is expressed as a percent. That is, a result of 10 = 10 percent.

Model objects currently supported: lm, glm, nls, betareg, gls, lme, lmer, lmerTest, glmmTMB, rq, loess, gam, glm.nb, glmRob, mblm, and rlm.

Value

A list of two objects: The series of model calls, and a data frame of statistics for each model.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/G_14.html

See Also

compareLM, compareGLM, nagelkerke

Examples

data(BrendonSmall)
BrendonSmall$Calories = as.numeric(BrendonSmall$Calories)
BrendonSmall$Calories2 = BrendonSmall$Calories ^ 2
model.1 = lm(Sodium ~ Calories, data = BrendonSmall)

accuracy(model.1, plotit=FALSE)

model.2 = lm(Sodium ~ Calories + Calories2, data = BrendonSmall)
model.3 = glm(Sodium ~ Calories, data = BrendonSmall, family="Gamma")
quadplat = function(x, a, b, clx) {
          ifelse(x  < clx, a + b * x   + (-0.5*b/clx) * x   * x,
                           a + b * clx + (-0.5*b/clx) * clx * clx)}
model.4 = nls(Sodium ~ quadplat(Calories, a, b, clx),
              data = BrendonSmall,
              start = list(a=519, b=0.359, clx = 2300))
              
accuracy(list(model.1, model.2, model.3, model.4), plotit=FALSE)

### Perfect and poor model fits
X = c(1, 2,  3,  4,  5,  6, 7, 8, 9, 10, 11, 12)
Y = c(1, 2,  3,  4,  5,  6, 7, 8, 9, 10, 11, 12)
Z = c(1, 12, 13, 6, 10, 13, 4, 3, 5,  6, 10, 14)
perfect = lm(Y ~ X)
poor    = lm(Z ~ X)
accuracy(list(perfect, poor), plotit=FALSE)

Description

A matrix of counts for students passing or failing a pesticide training course across four counties. Hypothetical data.

Usage

Anderson

Format

An object of class matrix (inherits from array) with 4 rows and 2 columns.

Source

https://rcompanion.org/handbook/H_04.html

Description

A data frame of counts for students passing or failing a pesticicde training course across four counties, with gender of students. Hypothetical data.

Usage

AndersonBias

Format

An object of class data.frame with 16 rows and 4 columns.

Source

https://rcompanion.org/handbook/H_06.html

Description

A matrix of paired counts for students planning to install rain barrels before and after a class. Hypothetical data.

Usage

AndersonRainBarrel

Format

An object of class matrix (inherits from array) with 2 rows and 2 columns.

Source

https://rcompanion.org/handbook/H_05.html

Description

A matrix of paired counts for students planning to install rain gardens before and after a class. Hypothetical data.

Usage

AndersonRainGarden

Format

An object of class matrix (inherits from array) with 3 rows and 3 columns.

Source

https://rcompanion.org/handbook/H_05.html

Description

Normal scores transformation (Inverse normal transformation) by Elfving, Blom, van der Waerden, Tukey, and rankit methods, as well as z score transformation (standardization) and scaling to a range (normalization).

Usage

blom(
  x,
  method = "general",
  alpha = pi/8,
  complete = FALSE,
  na.last = "keep",
  na.rm = TRUE,
  adjustN = TRUE,
  min = 1,
  max = 10,
  ...
)

Arguments

Details

By default, NA values are retained in the output. This behavior can be changed with the na.rm argument for "zscore" and "scale" methods, or with na.last for the normal scores methods. Or NA values can be removed from the input with complete=TRUE.

For normal scores methods, if there are NA values or tied values, it is helpful to look up the documentation for rank.

In general, for normal scores methods, either of the arguments method or alpha can be used. With the current algorithms, there is no need to use both.

Normal scores transformation will return a normal distribution with a mean of 0 and a standard deviation of 1.

The "scale" method coverts values to the range specified in max and min without transforming the distribution of values. By default, the "scale" method converts values to a 1 to 10 range. Using the "scale" method with min = 0 and max = 1 is sometimes called "normalization".

The "zscore" method converts values by the usual method for z scores: (x - mean(x)) / sd(x). The transformed values with have a mean of 0 and a standard deviation of 1 but won't be coerced into a normal distribution. Sometimes this method is called "standardization".

Value

A vector of numeric values.

Note

It's possible that Gustav Elfving didn't recommend the formula used in this function for the Elfving method. I would like thank Terence Cooke at the University of Exeter for their diligence at trying to track down a reference for this formula.

Author(s)

Salvatore Mangiafico, [email protected]

References

Conover, 1995, Practical Nonparametric Statistics, 3rd.

Solomon & Sawilowsky, 2009, Impact of rank-based normalizing transformations on the accuracy of test scores.

Beasley and Erickson, 2009, Rank-based inverse normal transformations are increasingly used, but are they merited?

Examples

set.seed(12345)
A = rlnorm(100)
## Not run: hist(A)
### Convert data to normal scores by Elfving method
B = blom(A)
## Not run: hist(B)
### Convert data to z scores 
C = blom(A, method="zscore")
## Not run: hist(C)
### Convert data to a scale of 1 to 10 
D = blom(A, method="scale")
## Not run: hist(D)

### Data from Sokal and Rohlf, 1995, 
### Biometry: The Principles and Practice of Statistics
### in Biological Research
Value = c(709,679,699,657,594,677,592,538,476,508,505,539)
Sex   = c(rep("Male",3), rep("Female",3), rep("Male",3), rep("Female",3))
Fat   = c(rep("Fresh", 6), rep("Rancid", 6))
ValueBlom = blom(Value)
Sokal = data.frame(ValueBlom, Sex, Fat)
model = lm(ValueBlom ~ Sex * Fat, data=Sokal)
anova(model)
## Not run: 
hist(residuals(model))
plot(predict(model), residuals(model))

## End(Not run)

Description

A data frame of Likert responses for five instructors for each of 8 respondents. Arranged in unreplicated complete block design. Hypothetical data.

Usage

BobBelcher

Format

An object of class data.frame with 40 rows and 3 columns.

Source

https://rcompanion.org/handbook/F_10.html

Description

A two-dimensional contingency table, in which Breakfast is an ordered nominal variable, and Travel is a non-ordered nominal variable. Hypothetical data.

Usage

Breakfast

Format

An object of class table with 3 rows and 5 columns.

Source

https://rcompanion.org/handbook/H_09.html

Description

A data frame of the intake of calories and sodium for students in five classes. Hypothetical data.

Usage

BrendonSmall

Format

An object of class data.frame with 45 rows and 6 columns.

Source

https://rcompanion.org/handbook/I_10.html

Description

A data frame of counts of students passing and failing. Hypothetical data.

Usage

BullyHill

Format

An object of class data.frame with 12 rows and 5 columns.

Source

https://rcompanion.org/handbook/J_02.html

Description

A data frame of the number of steps taken by students in three classes. Hypothetical data.

Usage

Catbus

Format

An object of class data.frame with 26 rows and 5 columns.

Source

https://rcompanion.org/handbook/C_03.html

Description

Produces critical-x and critical-y values for bivariate data according to a Cate-Nelson analysis.

Usage

cateNelson(
  x,
  y,
  plotit = TRUE,
  hollow = TRUE,
  xlab = "X",
  ylab = "Y",
  trend = "positive",
  clx = 1,
  cly = 1,
  xthreshold = 0.1,
  ythreshold = 0.1,
  progress = TRUE,
  verbose = TRUE,
  listout = FALSE
)

Arguments

Details

Cate-Nelson analysis divides bivariate data into two groups. For data with a positive trend, one group has a large x value associated with a large y value, and the other group has a small x value associated with a small y value. For a negative trend, a small x is associated with a large y, and so on.

The analysis is useful for bivariate data which don't conform well to linear, curvilinear, or plateau models.

This function will fail if either of the largest two or smallest two x values are identical.

Value

A data frame of statistics from the analysis: number of observations, critical level for x, sum of squares, critical value for y, the number of observations in each of the quadrants (I, II, III, IV), the number of observations that conform with the model, the proportion of observations that conform with the model, the number of observations that do not conform to the model, the proportion of observations that do not conform to the model, a p-value for the Fisher exact test for the data divided into the groups indicated by the model, and Cramer's V for the data divided into the groups indicated by the model.

Output also includes printed lists of critical values, explanation of the values in the data frame, and plots: y vs. x; sum of squares vs. critical x value; the number of observations that do not conform to the model vs. critical y value; and y vs. x with the critical values shown as lines on the plot, and the quadrants labeled.

Note

The method in this function follows Cate, R. B., & Nelson, L.A. (1971). A simple statistical procedure for partitioning soil test correlation data into two classes. Soil Science Society of America Proceedings 35, 658-660.

An earlier version of this function was published in Mangiafico, S.S. 2013. Cate-Nelson Analysis for Bivariate Data Using R-project. J.of Extension 51:5, 5TOT1.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/rcompanion/h_02.html

Cate, R. B., & Nelson, L.A. (1971). A simple statistical procedure for partitioning soil test correlation data into two classes. Soil Science Society of America Proceedings 35, 658–660.

See Also

cateNelsonFixedY

Examples

data(Nurseries)
cateNelson(x          = Nurseries$Size,
           y          = Nurseries$Proportion,
           plotit     = TRUE,
           hollow     = TRUE,
           xlab       = "Nursery size in hectares",
           ylab       = "Proportion of good practices adopted",
           trend      = "positive",
           clx        = 1,
           xthreshold = 0.10,
           ythreshold = 0.15)

Description

Produces critical-x values for bivariate data according to a Cate-Nelson analysis for a given critical Y value.

Usage

cateNelsonFixedY(
  x,
  y,
  cly = 0.95,
  plotit = TRUE,
  hollow = TRUE,
  xlab = "X",
  ylab = "Y",
  trend = "positive",
  clx = 1,
  outlength = 20,
  sortstat = "error"
)

Arguments

Details

Cate-Nelson analysis divides bivariate data into two groups. For data with a positive trend, one group has a large x value associated with a large y value, and the other group has a small x value associated with a small y value. For a negative trend, a small x is associated with a large y, and so on.

The analysis is useful for bivariate data which don't conform well to linear, curvilinear, or plateau models.

Value

A data frame of statistics from the analysis: critical level for x, critical value for y, the number of observations in each of the quadrants (I, II, III, IV), the number of observations that conform with the model, the number of observations that do not conform to the model, the proportion of observations that conform with the model, the proportion of observations that do not conform to the model, a p-value for the Fisher exact test for the data divided into the groups indicated by the model, phi for the data divided into the groups indicated by the model, and Pearson's chi-square for the data divided into the groups indicated by the model.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/rcompanion/h_02.html

See Also

cateNelson

Examples

data(Nurseries)
cateNelsonFixedY(x          = Nurseries$Size,
                 y          = Nurseries$Proportion,
                 cly        = 0.70,
                 plotit     = TRUE,
                 hollow     = TRUE,
                 xlab       = "Nursery size in hectares",
                 ylab       = "Proportion of good practices adopted",
                 trend      = "positive",
                 clx        = 1,
                 outlength  = 15)

Description

Produces a compact letter display (cld) from pairwise comparisons that were summarized in a table of comparisons

Usage

cldList(
  formula = NULL,
  data = NULL,
  comparison = NULL,
  p.value = NULL,
  threshold = 0.05,
  print.comp = FALSE,
  remove.space = TRUE,
  remove.equal = TRUE,
  remove.zero = TRUE,
  swap.colon = TRUE,
  swap.vs = FALSE,
  ...
)

Arguments

Details

The input should include either formula and data; or comparison and p.value.

This function relies upon the multcompLetters function in the multcompView package. The text for the comparisons passed to multcompLetters should be in the form "Treat.A-Treat.B". Currently by default cldList removes spaces, equal signs, and zeros, by default, and so can use text in the form e.g. "Treat.A - Treat.B = 0". It also changes ":" to "-", and so can use text in the form e.g. "Treat.A : Treat.B".

Value

A data frame of group names, group separation letters, and monospaced separtions letters

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The first variable on the right side is used for the grouping variable.

It is often helpful to reorder the factor levels in the data set so that the group with the largest e.g. mean or median is first, and so on.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/G_06.html

See Also

fullPTable

Examples

data(BrendonSmall)

model = aov(Calories ~ Instructor, data=BrendonSmall)

TUK = TukeyHSD(model, "Instructor", ordered = TRUE)

### Convert the TukeyHSD output to a standard data frame

TUK = as.data.frame(TUK$Instructor)
names(TUK) = gsub(" ", ".", names(TUK))

HSD = data.frame(Comparison=row.names(TUK), 
                 diff=TUK$diff, lwr=TUK$lwr, lwr=TUK$lwr, p.adj=TUK$p.adj)

HSD

cldList(p.adj ~ Comparison, data = HSD,
        threshold = 0.05,
        remove.space=FALSE)

Description

Calculates Cliff's delta with confidence intervals by bootstrap

Usage

cliffDelta(
  formula = NULL,
  data = NULL,
  x = NULL,
  y = NULL,
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  reportIncomplete = FALSE,
  brute = FALSE,
  verbose = FALSE,
  digits = 3,
  ...
)

Arguments

Details

Cliff's delta is an effect size statistic appropriate in cases where a Wilcoxon-Mann-Whitney test might be used. It ranges from -1 to 1, with 0 indicating stochastic equality, and 1 indicating that the first group dominates the second. It is linearly related to Vargha and Delaney's A.

By default, the function calculates Cliff's delta from the "W" U statistic from the wilcox.test function. Specifically, VDA = U/(n1*n2); CD = (VDA-0.5)*2.

The input should include either formula and data; or x, and y. If there are more than two groups, only the first two groups are used.

Currently, the function makes no provisions for NA values in the data. It is recommended that NAs be removed beforehand.

When the data in the first group are greater than in the second group, Cliff's delta is positive. When the data in the second group are greater than in the first group, Cliff's delta is negative.

Be cautious with this interpretation, as R will alphabetize groups in the formula interface if the grouping variable is not already a factor.

When Cliff's delta is close to 1 or close to -1, or with small sample size, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

Value

A single statistic, Cliff's delta. Or a small data frame consisting of Cliff's delta, and the lower and upper confidence limits.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The first variable on the right side is used for the grouping variable

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_04.html

See Also

vda, multiVDA

Examples

data(Catbus)
cliffDelta(Steps ~ Gender, data=Catbus)

Description

Calculates Cohen's g and odds ratio for paired contingency tables, such as those that might be analyzed with McNemar or McNemar-Bowker tests.

Usage

cohenG(
  x,
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 3,
  reportIncomplete = FALSE,
  ...
)

Arguments

Details

For a 2 x 2 table, where a and d are the concordant cells and b and c are discordant cells: Odds ratio is b/c; P is b/(b+c); and Cohen's g is P - 0.5.

In the 2 x 2 case, the statistics are directional. That is, when cell [1, 2] in the table is greater than cell [2, 1], OR is greater than 1, P is greater than 0.5, and g is positive.

In the opposite case, OR is less than 1, P is less than 0.5, and g is negative.

In the 2 x 2 case, when the effect is small, the confidence interval for OR can pass through 1, for g can pass through 0, and for P can pass through 0.5.

For tables larger than 2 x 2, the statistics are not directional. That is, OR is always >= 1, P is always >= 0.5, and g is always positive. Because of this, if type="perc", the confidence interval will never cross the values for no effect (OR = 1, P = 0.5, or g = 0). Because of this, the confidence interval range in this case should not be used for statistical inference. However, if type="norm", the confidence interval may cross the values for no effect.

When the reported statistics are close to their extremes, or with small counts, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

Value

A list containing: a data frame of results of the global statistics; and a data frame of results of the pairwise statistics.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/H_05.html

See Also

nominalSymmetryTest, cohenH

Examples

### 2 x 2 repeated matrix example
data(AndersonRainBarrel)
cohenG(AndersonRainBarrel)
                    
### 3 x 3 repeated matrix
data(AndersonRainGarden)
cohenG(AndersonRainGarden)

Description

Calculates Cohen's h for 2 x 2 contingency tables, such as those that might be analyzed with a chi-square test of association.

Usage

cohenH(x, observation = "row", verbose = TRUE, digits = 3)

Arguments

Details

Cohen's h is an effect size to compare two proportions. For a 2 x 2 table: Cohen's h equals Phi2 - Phi1, where, If observations are in rows, P1 = a/(a+b) and P2 = c/(c+d). If observations are in columns, P1 = a/(a+c) and P2 = b/(b+d). Phi = 2 * asin(sqrt(P))

Value

A single statistic.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/H_10.html

See Also

cohenG

Examples

data(Pennsylvania18)
Pennsylvania18
cohenH(Pennsylvania18, observation="row")

Description

Calculates Cohen's w for a table of nominal variables.

Usage

cohenW(
  x,
  y = NULL,
  p = NULL,
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 4,
  reportIncomplete = FALSE,
  ...
)

Arguments

Details

Cohen's w is used as a measure of association between two nominal variables, or as an effect size for a chi-square test of association. For a 2 x 2 table, the absolute value of the phi statistic is the same as Cohen's w. The value of Cohen's w is not bound by 1 on the upper end.

Cohen's w is "naturally nondirectional". That is, the value will always be zero or positive. Because of this, if type="perc", the confidence interval will never cross zero. The confidence interval range should not be used for statistical inference. However, if type="norm", the confidence interval may cross zero.

When w is close to 0 or very large, or with small counts, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

Value

A single statistic, Cohen's w. Or a small data frame consisting of Cohen's w, and the lower and upper confidence limits.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/H_10.html

Cohen J. 1992. "A Power Primer". Psychological Bulletin 12(1): 155-159.

Cohen, J. 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd Ed. Routledge.

See Also

cramerV

Examples

### Example with table
data(Anderson)
fisher.test(Anderson)
cohenW(Anderson)

### Example for goodness-of-fit
### Bird foraging example, Handbook of Biological Statistics
observed = c(70,   79,   3,    4)
expected = c(0.54, 0.40, 0.05, 0.01)
chisq.test(observed, p = expected)
cohenW(observed, p = expected)

### Example with two vectors
Species = c(rep("Species1", 16), rep("Species2", 16))
Color   = c(rep(c("blue", "blue", "blue", "green"),4),
            rep(c("green", "green", "green", "blue"),4))
fisher.test(Species, Color)
cohenW(Species, Color)

Description

Produces a table of fit statistics for multiple glm models.

Usage

compareGLM(fits, ...)

Arguments

Details

Produces a table of fit statistics for multiple glm models: AIC, AICc, BIC, p-value, pseudo R-squared (McFadden, Cox and Snell, Nagelkerke).

Smaller values for AIC, AICc, and BIC indicate a better balance of goodness-of-fit of the model and the complexity of the model. The goal is to find a model that adequately explains the data without having too many terms.

BIC tends to choose models with fewer parameters relative to AIC.

For comparisons with AIC, etc., to be valid, both models must have the same data, without transformations, use the same dependent variable, and be fit with the same method. They do not need to be nested.

The function will fail if a model formula is longer than 500 characters.

Value

A list of two objects: The series of model calls, and a data frame of statistics for each model.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/rcompanion/e_07.html

See Also

compareLM, pairwiseModelAnova, accuracy

Examples

### Compare among logistic regresion models
data(AndersonBias)
model.0 = glm(Result ~ 1, weight = Count, data = AndersonBias,
             family = binomial(link="logit"))
model.1 = glm(Result ~ County, weight = Count, data = AndersonBias,
             family = binomial(link="logit"))
model.2 = glm(Result ~ County + Gender, weight = Count, data = AndersonBias,
             family = binomial(link="logit"))
model.3 = glm(Result ~ County + Gender + County:Gender, weight = Count, 
             data = AndersonBias, family = binomial(link="logit"))
compareGLM(model.0, model.1, model.2, model.3)

Description

Produces a table of fit statistics for multiple lm models.

Usage

compareLM(fits, ...)

Arguments

Details

Produces a table of fit statistics for multiple lm models: AIC, AICc, BIC, p-value, R-squared, and adjusted R-squared.

Smaller values for AIC, AICc, and BIC indicate a better balance of goodness-of-fit of the model and the complexity of the model. The goal is to find a model that adequately explains the data without having too many terms.

BIC tends to choose models with fewer parameters relative to AIC.

In the table, Shapiro.W and Shapiro.p are the W statistic and p-value for the Shapiro-Wilks test on the residuals of the model.

For comparisons with AIC, etc., to be valid, both models must have the same data, without transformations, use the same dependent variable, and be fit with the same method. They do not need to be nested.

The function will fail if a model formula is longer than 500 characters.

Value

A list of two objects: The series of model calls, and a data frame of statistics for each model.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/I_10.html, https://rcompanion.org/rcompanion/e_05.html

See Also

compareGLM, pairwiseModelAnova, accuracy

Examples

### Compare among polynomial models
data(BrendonSmall)
BrendonSmall$Calories = as.numeric(BrendonSmall$Calories)

BrendonSmall$Calories2 = BrendonSmall$Calories * BrendonSmall$Calories
BrendonSmall$Calories3 = BrendonSmall$Calories * BrendonSmall$Calories * 
                         BrendonSmall$Calories
BrendonSmall$Calories4 = BrendonSmall$Calories * BrendonSmall$Calories * 
                         BrendonSmall$Calories * BrendonSmall$Calories
model.1 = lm(Sodium ~ Calories, data = BrendonSmall)
model.2 = lm(Sodium ~ Calories + Calories2, data = BrendonSmall)
model.3 = lm(Sodium ~ Calories + Calories2 + Calories3, data = BrendonSmall)
model.4 = lm(Sodium ~ Calories + Calories2 + Calories3 + Calories4,
             data = BrendonSmall)
compareLM(model.1, model.2, model.3, model.4)

Description

Produces measures of association for all variables in a data frame with confidence intervals when available.

Usage

correlation(
  data = NULL,
  printClasses = FALSE,
  progress = TRUE,
  methodNum = "pearson",
  methodOrd = "kendall",
  methodNumOrd = "spearman",
  methodNumNom = "eta",
  methodNumBin = "pearson",
  testChisq = "chisq",
  ci = FALSE,
  conf = 0.95,
  R = 1000,
  correct = FALSE,
  reportIncomplete = TRUE,
  na.action = "na.omit",
  digits = 3,
  pDigits = 4,
  ...
)

Arguments

Details

It’s important that variables are assigned the correct class to get an appropriate measure of association. That is, factor variables should be of class "factor", not "character". Ordered factors should be ordered factors (and have their levels in the correct order!).

Date variables are treated as numeric.

The default for measures of association tend to be "parametric" type. That is, e.g. Pearson correlation where appropriate.

Nonparametric measures of association will be reported with the options methodNum = "spearman", methodNumNom = "epsilon", methodNumBin = "glass".

Value

A data frame of variables, association statistics, p-values, and confidence intervals.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/I_14.html

See Also

phi, spearmanRho, cramerV, freemanTheta, wilcoxonRG

Examples

Length   = c(0.29, 0.25, NA, 0.40, 0.50, 0.57, 0.62, 0.88, 0.99, 0.90)
Rating   = factor(ordered=TRUE, levels=c("Low", "Medium", "High"),
                  x = rep(c("Low", "Medium", "High"), c(3,3,4)))
Color    = factor(rep(c("Red", "Green", "Blue"), c(4,4,2)))
Flag     = factor(rep(c(TRUE, FALSE, TRUE), c(5,4,1)))
Answer   = factor(rep(c("Yes", "No", "Yes"), c(4,3,3)), levels=c("Yes", "No"))
Location = factor(rep(c("Home", "Away", "Other"), c(2,4,4)))
Distance = factor(ordered=TRUE, levels=c("Low", "Medium", "High"),
                  x = rep(c("Low", "Medium", "High"), c(5,2,3))) 
Start    = seq(as.Date("2024-01-01"), by = "month", length.out = 10)
Data = data.frame(Length, Rating, Color, Flag, Answer, Location, Distance, Start)  
correlation(Data)

Description

Produces the count pseudo r-squared measure for models with a binary outcome.

Usage

countRSquare(
  fit,
  digits = 3,
  suppressWarnings = TRUE,
  plotit = FALSE,
  jitter = FALSE,
  pch = 1,
  ...
)

Arguments

Details

The count pseudo r-squared is simply the number of correctly predicted observations divided the total number of observations.

This version is appropriate for models with a binary outcome.

The adjusted value deducts the count of the most frequent outcome from both the numerator and the denominator.

It is recommended that the model is fit on data in long format. That is, that the weight option not be used in the model.

The function makes no provisions for NA values. It is recommended that NA values be removed before the determination of the model.

Value

A list including a description of the submitted model, a data frame with the pseudo r-squared results, and a confusion matrix of the results.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://stats.oarc.ucla.edu/other/mult-pkg/faq/general/faq-what-are-pseudo-r-squareds/, https://rcompanion.org/handbook/H_08.html, https://rcompanion.org/rcompanion/e_06.html

See Also

nagelkerke, efronRSquared, accuracy

Examples

data(AndersonBias)

### Covert data to long format

Long = AndersonBias[rep(row.names(AndersonBias), AndersonBias$Count),
                    c("Result", "County", "Gender")]
rownames(Long) = seq(1:nrow(Long))
str(Long)

### Fit model and determine count r-square

model = glm(Result ~ County + Gender + County:Gender,
            data = Long,
            family = binomial())

countRSquare(model)

Description

Calculates Cramer's V for a table of nominal variables; confidence intervals by bootstrap.

Usage

cramerV(
  x,
  y = NULL,
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 4,
  bias.correct = FALSE,
  reportIncomplete = FALSE,
  verbose = FALSE,
  tolerance = 1e-16,
  ...
)

Arguments

Details

Cramer's V is used as a measure of association between two nominal variables, or as an effect size for a chi-square test of association. For a 2 x 2 table, the absolute value of the phi statistic is the same as Cramer's V.

Because V is always positive, if type="perc", the confidence interval will never cross zero. In this case, the confidence interval range should not be used for statistical inference. However, if type="norm", the confidence interval may cross zero.

When V is close to 0 or very large, or with small counts, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

Value

A single statistic, Cramer's V. Or a small data frame consisting of Cramer's V, and the lower and upper confidence limits.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/H_10.html

See Also

phi, cohenW, cramerVFit

Examples

### Example with table
data(Anderson)
fisher.test(Anderson)
cramerV(Anderson)

### Example with two vectors
Species = c(rep("Species1", 16), rep("Species2", 16))
Color   = c(rep(c("blue", "blue", "blue", "green"),4),
            rep(c("green", "green", "green", "blue"),4))
fisher.test(Species, Color)
cramerV(Species, Color)

Description

Calculates Cramer's V for a vector of counts and expected counts; confidence intervals by bootstrap.

Usage

cramerVFit(
  x,
  p = rep(1/length(x), length(x)),
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 4,
  reportIncomplete = FALSE,
  verbose = FALSE,
  ...
)

Arguments

Details

This modification of Cramer's V could be used to indicate an effect size in cases where a chi-square goodness-of-fit test might be used. It indicates the degree of deviation of observed counts from the expected probabilities.

In the case of equally-distributed expected frequencies, Cramer's V will be equal to 1 when all counts are in one category, and it will be equal to 0 when the counts are equally distributed across categories. This does not hold if the expected frequencies are not equally-distributed.

Because V is always positive, if type="perc", the confidence interval will never cross zero, and should not be used for statistical inference. However, if type="norm", the confidence interval may cross zero.

When V is close to 0 or 1, or with small counts, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

In addition, the function will not return a confidence interval if there are zeros in any cell.

Value

A single statistic, Cramer's V. Or a small data frame consisting of Cramer's V, and the lower and upper confidence limits.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/H_03.html

See Also

cramerV

Examples

### Equal probabilities example
### From https://rcompanion.org/handbook/H_03.html
nail.color = c("Red", "None", "White", "Green", "Purple", "Blue")
observed   = c( 19,    3,      1,       1,       2,        2    )
expected   = c( 1/6,   1/6,    1/6,     1/6,     1/6,      1/6  )
chisq.test(x = observed, p = expected)
cramerVFit(x = observed, p = expected)

### Unequal probabilities example
### From https://rcompanion.org/handbook/H_03.html
race = c("White", "Black", "American Indian", "Asian", "Pacific Islander",
          "Two or more races")
observed = c(20, 9, 9, 1, 1, 1)
expected = c(0.775, 0.132, 0.012, 0.054, 0.002, 0.025)
chisq.test(x = observed, p = expected)
cramerVFit(x = observed, p = expected)

### Examples of perfect and zero fits
cramerVFit(c(100, 0, 0, 0, 0))
cramerVFit(c(10, 10, 10, 10, 10))

Description

Produces Efron's pseudo r-squared from certain models, or vectors of residuals, predicted values, and actual values. Alternately produces minimum maximum accuracy, mean absolute percent error, root mean square error, or coefficient of variation.

Usage

efronRSquared(
  model = NULL,
  actual = NULL,
  predicted = NULL,
  residual = NULL,
  statistic = "EfronRSquared",
  plotit = FALSE,
  digits = 3,
  ...
)

Arguments

Details

Efron's pseudo r-squared is calculated as 1 minus the residual sum of squares divided by the total sum of squares. For linear models (lm model objects), Efron's pseudo r-squared will be equal to r-squared.

This function produces the same statistics as does the accuracy function. While the accuracy function extracts values from a model object, this function allows for the manual entry of residual, predicted, or actual values.

It is recommended that the user consults the accuracy function for further details on these statistics, such as if the reported value is presented as a percentage or fraction.

If modelis not supplied, two of the following need to passed to the function: actual, predicted, residual.

Note that, for some model objects, to extract residuals and predicted values on the original scale, a type="response" option needs to be added to the call, e.g. residuals(model.object, type="response").

Value

A single statistic

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_16.html

See Also

accuracy, nagelkerke

Examples

data(BrendonSmall)
BrendonSmall$Calories = as.numeric(BrendonSmall$Calories)
BrendonSmall$Calories2 = BrendonSmall$Calories ^ 2
model.1 = lm(Sodium ~ Calories + Calories2, data = BrendonSmall)

efronRSquared(model.1)

efronRSquared(model.1, statistic="MAPE")

efronRSquared(actual=BrendonSmall$Sodium, residual=model.1$residuals)
efronRSquared(residual=model.1$residuals, predicted=model.1$fitted.values)
efronRSquared(actual=BrendonSmall$Sodium, predicted=model.1$fitted.values)

summary(model.1)$r.squared

Description

Calculates epsilon-squared as an effect size statistic, following a Kruskal-Wallis test, or for a table with one ordinal variable and one nominal variable; confidence intervals by bootstrap

Usage

epsilonSquared(
  x,
  g = NULL,
  group = "row",
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 3,
  reportIncomplete = FALSE,
  ...
)

Arguments

Details

Epsilon-squared is used as a measure of association for the Kruskal-Wallis test or for a two-way table with one ordinal and one nominal variable.

Currently, the function makes no provisions for NA values in the data. It is recommended that NAs be removed beforehand.

Because epsilon-squared is always positive, if type="perc", the confidence interval will never cross zero, and should not be used for statistical inference. However, if type="norm", the confidence interval may cross zero.

When epsilon-squared is close to 0 or very large, or with small counts in some cells, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

Value

A single statistic, epsilon-squared. Or a small data frame consisting of epsilon-squared, and the lower and upper confidence limits.

Note

Note that epsilon-squared as calculated by this function is equivalent to the eta-squared, or r-squared, as determined by an anova on the rank-transformed values. Epsilon-squared for Kruskal-Wallis is typically defined this way in the literature.

Author(s)

Salvatore Mangiafico, [email protected]

References

King, B.M., P.J. Rosopa, and E.W. Minium. 2018. Statistical Reasoning in the Behavioral Sciences, 7th ed. Wiley.

https://rcompanion.org/handbook/F_08.html

See Also

multiVDA, ordinalEtaSquared

Examples

data(Breakfast)
library(coin)
chisq_test(Breakfast, scores = list("Breakfast" = c(-2, -1, 0, 1, 2)))
epsilonSquared(Breakfast)

data(PoohPiglet)
kruskal.test(Likert ~ Speaker, data = PoohPiglet)
epsilonSquared(x = PoohPiglet$Likert, g = PoohPiglet$Speaker)

### Same data, as matrix of counts
data(PoohPiglet)
XT = xtabs( ~ Speaker + Likert , data = PoohPiglet)
epsilonSquared(XT)

Description

Calculates Freeman's theta for a table with one ordinal variable and one nominal variable; confidence intervals by bootstrap.

Usage

freemanTheta(
  x,
  g = NULL,
  group = "row",
  verbose = FALSE,
  progress = FALSE,
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 3,
  reportIncomplete = FALSE
)

Arguments

Details

Freeman's coefficent of differentiation (theta) is used as a measure of association for a two-way table with one ordinal and one nominal variable. See Freeman (1965).

Currently, the function makes no provisions for NA values in the data. It is recommended that NAs be removed beforehand.

Because theta is always positive, if type="perc", the confidence interval will never cross zero, and should not be used for statistical inference. However, if type="norm", the confidence interval may cross zero.

When theta is close to 0 or very large, or with small counts in some cells, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

Value

A single statistic, Freeman's theta. Or a small data frame consisting of Freeman's theta, and the lower and upper confidence limits.

Author(s)

Salvatore Mangiafico, [email protected]

References

Freeman, L.C. 1965. Elementary Applied Statistics for Students in Behavioral Science. Wiley.

https://rcompanion.org/handbook/H_11.html

See Also

epsilonSquared

Examples

data(Breakfast)
library(coin)
chisq_test(Breakfast, scores = list("Breakfast" = c(-2, -1, 0, 1, 2)))
freemanTheta(Breakfast)

### Example from Freeman (1965), Table 10.6
Counts = c(1, 2, 5, 2, 0, 10, 5, 5, 0, 0, 0, 0, 2, 2, 1, 0, 0, 0, 2, 3)
Matrix = matrix(Counts, byrow=TRUE, ncol=5,
                dimnames = list(Marital.status = c("Single", "Married",
                                                   "Widowed", "Divorced"),
                                Social.adjustment = c("5","4","3","2","1")))
Matrix
freemanTheta(Matrix)

### Example after Kruskal Wallis test
data(PoohPiglet)
kruskal.test(Likert ~ Speaker, data = PoohPiglet)
freemanTheta(x = PoohPiglet$Likert, g = PoohPiglet$Speaker)

### Same data, as table of counts
data(PoohPiglet)
XT = xtabs( ~ Speaker + Likert , data = PoohPiglet)
freemanTheta(XT)

### Example from Freeman (1965), Table 10.7
Counts = c(52, 28, 40, 34, 7, 9, 16, 10, 8, 4, 10, 9, 12,6, 7, 5)
Matrix = matrix(Counts, byrow=TRUE, ncol=4,
                dimnames = list(Preferred.trait = c("Companionability",
                                                    "PhysicalAppearance",
                                                    "SocialGrace",
                                                    "Intelligence"),
                                Family.income = c("4", "3", "2", "1")))
Matrix
freemanTheta(Matrix, verbose=TRUE)

Description

Converts a lower triangle matrix to a full matrix.

Usage

fullPTable(PT)

Arguments

Details

This function is useful to convert a lower triangle matrix of p-values from a pairwise test to a full matrix. A full matrix can be passed to multcompLetters in the multcompView package to produce a compact letter display.

Value

A full matrix.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_08.html

See Also

cldList

Examples

### Example with pairwise.wilcox.test
data(BrendonSmall)
BrendonSmall$Instructor = factor(BrendonSmall$Instructor,
                          levels = c('Brendon Small', 'Jason Penopolis',
                                     'Paula Small', 'Melissa Robbins', 
                                     'Coach McGuirk'))
P   = pairwise.wilcox.test(x = BrendonSmall$Score, g = BrendonSmall$Instructor)
PT  = P$p.value
PT
PT1 = fullPTable(PT)
PT1
library(multcompView)
multcompLetters(PT1)

Description

Conducts groupwise tests of association on a three-way contingency table.

Usage

groupwiseCMH(
  x,
  group = 3,
  fisher = TRUE,
  gtest = FALSE,
  chisq = FALSE,
  method = "fdr",
  correct = "none",
  digits = 3,
  ...
)

Arguments

Details

If more than one of fisher, gtest, or chisq is set to TRUE, only one type of test of association will be conducted.

Value

A data frame of groups, test used, p-values, and adjusted p-values.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/H_06.html

See Also

nominalSymmetryTest, pairwiseMcnemar, pairwiseNominalIndependence, pairwiseNominalMatrix

Examples

### Post-hoc for Cochran-Mantel-Haenszel test
data(AndersonBias)
Table = xtabs(Count ~ Gender + Result + County,
              data=AndersonBias)
ftable(Table)
mantelhaen.test(Table)
groupwiseCMH(Table,
             group   = 3,
             fisher  = TRUE,
             gtest   = FALSE,
             chisq   = FALSE,
             method  = "fdr",
             correct = "none",
             digits  = 3)

Description

Calculates geometric means and confidence intervals for groups.

Usage

groupwiseGeometric(
  formula = NULL,
  data = NULL,
  var = NULL,
  group = NULL,
  conf = 0.95,
  na.rm = TRUE,
  digits = 3,
  ...
)

Arguments

Details

The input should include either formula and data; or data, var, and group. (See examples).

The function computes means, standard deviations, standard errors, and confidence intervals on log-transformed values. Confidence intervals are calculated in the traditional manner with the t-distribution on the transformed values, and then back-transforms the confidence interval limits. These statistics assume that the data are log-normally distributed. For data not meeting this assumption, medians and confidence intervals by bootstrap may be more appropriate.

Value

A data frame of geometric means, standard deviations, standard errors, and confidence intervals.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The variables on the right side are used for the grouping variables.

Results for ungrouped (one-sample) data can be obtained by either setting the right side of the formula to 1, e.g. y ~ 1, or by setting group=NULL.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/C_03.html

See Also

groupwiseMean, groupwiseMedian, groupwiseHuber

Examples

### Example with formula notation 
data(Catbus)
groupwiseGeometric(Steps ~ Gender + Teacher,
                   data   = Catbus)

### Example with variable notation                                              
data(Catbus)
groupwiseGeometric(data   = Catbus,
                   var    = "Steps",
                   group  = c("Gender", "Teacher"))

Description

Calculates Huber M-estimator and confidence intervals for groups.

Usage

groupwiseHuber(
  formula = NULL,
  data = NULL,
  var = NULL,
  group = NULL,
  conf.level = 0.95,
  ci.type = "wald",
  digits = 3,
  ...
)

Arguments

Details

A wrapper for the DescTools::HuberM function to allow easy output for multiple groups.

The input should include either formula and data; or data, var, and group. (See examples).

Results for ungrouped (one-sample) data can be obtained by either setting the right side of the formula to 1, e.g. y ~ 1, or by setting group=NULL.

Value

A data frame of requested statistics by group.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The variables on the right side are used for the grouping variables.

It is recommended to remove NA values before using this function. At the time of writing, NA values will cause the function to fail if confidence intervals are requested.

At the time of writing, the ci.type="boot" option produces NA results. This is a result from the DescTools::HuberM function.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/rcompanion/d_08a.html

See Also

groupwiseMean, groupwiseMedian, groupwiseGeometric

Examples

### Example with formula notation
data(Catbus)
groupwiseHuber(Steps ~ Teacher + Gender,
               data      = Catbus,
               ci.type   = "wald")
               
### Example with variable notation
data(Catbus)
groupwiseHuber(data      = Catbus,
               var       = "Steps",
               group     = c("Teacher", "Gender"),
               ci.type   = "wald")

### Example with NA value and without confidence intervals
data(Catbus)
Catbus1 = Catbus
Catbus1[1, 'Steps'] = NA
groupwiseHuber(Steps ~ Teacher + Gender,
               data      = Catbus1,
               conf.level   = NA)

Description

Calculates means and confidence intervals for groups.

Usage

groupwiseMean(
  formula = NULL,
  data = NULL,
  var = NULL,
  group = NULL,
  trim = 0,
  na.rm = FALSE,
  conf = 0.95,
  R = 5000,
  boot = FALSE,
  traditional = TRUE,
  normal = FALSE,
  basic = FALSE,
  percentile = FALSE,
  bca = FALSE,
  digits = 3,
  ...
)

Arguments

Details

The input should include either formula and data; or data, var, and group. (See examples).

Results for ungrouped (one-sample) data can be obtained by either setting the right side of the formula to 1, e.g. y ~ 1, or by setting group=NULL when using var.

Value

A data frame of requested statistics by group.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The variables on the right side are used for the grouping variables.

In general, it is advisable to handle NA values before using this function. With some options, the function may not handle missing values well, or in the manner desired by the user. In particular, if bca=TRUE and there are NA values, the function may fail.

For a traditional method to calculate confidence intervals on trimmed means, see Rand Wilcox, Introduction to Robust Estimation and Hypothesis Testing.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/C_03.html

See Also

groupwiseMedian, groupwiseHuber, groupwiseGeometric

Examples

### Example with formula notation
data(Catbus)
groupwiseMean(Steps ~ Teacher + Gender,
              data        = Catbus,
              traditional = FALSE,
              percentile  = TRUE)

### Example with variable notation
data(Catbus)
groupwiseMean(data        = Catbus,
              var         = "Steps",
              group       = c("Teacher", "Gender"),
              traditional = FALSE,
              percentile  = TRUE)

Description

Calculates medians and confidence intervals for groups.

Usage

groupwiseMedian(
  formula = NULL,
  data = NULL,
  var = NULL,
  group = NULL,
  conf = 0.95,
  R = 5000,
  boot = FALSE,
  pseudo = FALSE,
  basic = FALSE,
  normal = FALSE,
  percentile = FALSE,
  bca = TRUE,
  wilcox = FALSE,
  exact = FALSE,
  digits = 3,
  ...
)

Arguments

Details

The input should include either formula and data; or data, var, and group. (See examples).

With some options, the function may not handle missing values well. This seems to happen particularly with bca = TRUE.

Value

A data frame of requested statistics by group.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The variables on the right side are used for the grouping variables.

Results for ungrouped (one-sample) data can be obtained by either setting the right side of the formula to 1, e.g. y ~ 1, or by setting group=NULL.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/E_04.html

See Also

groupwiseMean, groupwiseHuber, groupwiseGeometric

Examples

### Example with formula notation
data(Catbus)
groupwiseMedian(Steps ~ Teacher + Gender,
                data        = Catbus,
                bca         = FALSE,
                percentile  = TRUE,
                R           = 1000)
                
### Example with variable notation
data(Catbus)
groupwiseMedian(data         = Catbus,
                var         = "Steps",
                group       = c("Teacher", "Gender"),
                bca         = FALSE,
                percentile  = TRUE,
                R           = 1000)

Description

Calculates percentiles and confidence intervals for groups.

Usage

groupwisePercentile(
  formula = NULL,
  data = NULL,
  var = NULL,
  group = NULL,
  conf = 0.95,
  tau = 0.5,
  type = 7,
  R = 5000,
  boot = FALSE,
  basic = FALSE,
  normal = FALSE,
  percentile = FALSE,
  bca = TRUE,
  digits = 3,
  ...
)

Arguments

Details

The input should include either formula and data; or data, var, and group. (See examples).

With some options, the function may not handle missing values well. This seems to happen particularly with bca = TRUE.

Value

A data frame of requested statistics by group

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The variables on the right side are used for the grouping variables.

Results for ungrouped (one-sample) data can be obtained by either setting the right side of the formula to 1, e.g. y ~ 1, or by setting group=NULL.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_15.html

See Also

groupwiseMean, groupwiseHuber, groupwiseGeometric, groupwiseMedian

Examples

### Example with formula notation
data(Catbus)
groupwisePercentile(Steps ~ Teacher + Gender,
                    data        = Catbus,
                    tau         = 0.25,
                    bca         = FALSE,
                    percentile  = TRUE,
                    R           = 1000)
                
### Example with variable notation
data(Catbus)
groupwisePercentile(data         = Catbus,
                    var         = "Steps",
                    group       = c("Teacher", "Gender"),
                    tau         = 0.25,
                    bca         = FALSE,
                    percentile  = TRUE,
                    R           = 1000)

Description

Calculates sums for groups.

Usage

groupwiseSum(
  formula = NULL,
  data = NULL,
  var = NULL,
  group = NULL,
  digits = NULL,
  ...
)

Arguments

Details

The input should include either formula and data; or data, var, and group. (See examples).

Value

A data frame of statistics by group.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The variables on the right side are used for the grouping variables.

Beginning in version 2.0, there is no rounding of results by default. Rounding results can cause confusion if the user is expecting exact sums.

Author(s)

Salvatore Mangiafico, [email protected]

See Also

groupwiseMean, groupwiseMedian, groupwiseHuber, groupwiseGeometric

Examples

### Example with formula notation
data(AndersonBias)
groupwiseSum(Count ~ Result + Gender,
             data        = AndersonBias)
                
### Example with variable notation
data(AndersonBias)
groupwiseSum(data        = AndersonBias,
             var         = "Count",
             group       = c("Result", "Gender"))

Description

A data frame in long form with yes/no responses for four lawn care practices for each of 14 respondents. Hypothetical data.

Usage

HayleySmith

Format

An object of class data.frame with 56 rows and 3 columns.

Source

https://rcompanion.org/handbook/H_05.html

Description

Calculates Kendall's W coefficient of concordance, which can be used as an effect size statistic for unreplicated complete block design such as where Friedman's test might be used. This function is a wrapper for the KendallW function in the DescTools package, with the addition of bootstrapped confidence intervals.

Usage

kendallW(
  x,
  correct = TRUE,
  na.rm = FALSE,
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 3,
  ...
)

Arguments

Details

See the KendallW function in the DescTools package for details.

When W is close to 0 or very large, or with small sample size, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

Because W is always positive, if type="perc", the confidence interval will never cross zero, and should not be used for statistical inference. However, if type="norm", the confidence interval may cross zero.

When producing confidence intervals by bootstrap, this function treats each rater or block as an observation. It is not clear to the author if this approach produces accurate confidence intervals, but it appears to be reasonable.

Value

A single statistic, W. Or a small data frame consisting of W, and the lower and upper confidence limits.

Acknowledgments

My thanks to Indrajeet Patil, author of ggstatsplot, and groupedstats for help in the inspiring and coding of this function.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_10.html

Examples

data(BobBelcher)
Table = xtabs(Likert ~ Instructor + Rater, data = BobBelcher)
kendallW(Table)

Description

Calculates Mangiafico's d, which is the difference in medians divided by the pooled median absolute deviation, with confidence intervals by bootstrap

Usage

mangiaficoD(
  formula = NULL,
  data = NULL,
  x = NULL,
  y = NULL,
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  reportIncomplete = FALSE,
  verbose = FALSE,
  digits = 3,
  ...
)

Arguments

Details

Mangiafico's d is an appropriate effect size statistic where Mood's median test, or another test comparing two medians, might be used. Note that the response variable is treated as at least interval.

For normal samples, the result will be somewhat similar to Cohen's d.

The input should include either formula and data; or x, and y. If there are more than two groups, only the first two groups are used.

Currently, the function makes no provisions for NA values in the data. It is recommended that NAs be removed beforehand.

When the data in the first group are greater than in the second group, d is positive. When the data in the second group are greater than in the first group, d is negative.

Be cautious with this interpretation, as R will alphabetize groups in the formula interface if the grouping variable is not already a factor.

When d is close to 0 or close to 1, or with small sample size, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

Value

A single statistic, d. Or a small data frame consisting of d, and the lower and upper confidence limits.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The first variable on the right side is used for the grouping variable.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_05.html

See Also

multiMangiaficoD

Examples

data(Catbus)
mangiaficoD(Steps ~ Gender, data=Catbus, verbose=TRUE)

Nadja = c(5,5,6,6,6,7,7,11,11,11)
Nandor = c(0,1,2,3,4,5,6,7,8,9,10,11)
mangiaficoD(x = Nadja, y = Nandor, verbose=TRUE)

Description

A data frame of the number of monarch butterflies in three gardens. Hypothetical data.

Usage

Monarchs

Format

An object of class data.frame with 24 rows and 2 columns.

Source

https://rcompanion.org/handbook/J_01.html

Description

Calculates Mangiafico's d, which is the difference in medians divided by the pooled median absolute deviation, for several groups in a pairwise manner.

Usage

multiMangiaficoD(
  formula = NULL,
  data = NULL,
  x = NULL,
  g = NULL,
  digits = 3,
  ...
)

Arguments

Details

Mangiafico's d is an appropriate effect size statistic where Mood's median test, or another test comparing two medians, might be used. Note that the response variable is treated as at least interval.

When the data in the first group are greater than in the second group, d is positive. When the data in the second group are greater than in the first group, d is negative.

Be cautious with this interpretation, as R will alphabetize groups in the formula interface if the grouping variable is not already a factor.

Currently, the function makes no provisions for NA values in the data. It is recommended that NAs be removed beforehand.

Value

A list containing a data frame of pairwise statistics, and the comparison with the most extreme value of the statistic.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The first variable on the right side is used for the grouping variable.

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The first variable on the right side is used for the grouping variable.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_09.html

See Also

mangiaficoD

Examples

data(Catbus)
multiMangiaficoD(Steps ~ Teacher, data=Catbus)

Description

Calculates Vargha and Delaney's A (VDA), Cliff's delta (CD), and the Glass rank biserial coefficient, rg, for several groups in a pairwise manner.

Usage

multiVDA(
  formula = NULL,
  data = NULL,
  x = NULL,
  g = NULL,
  statistic = "VDA",
  digits = 3,
  ...
)

Arguments

Details

VDA and CD are effect size statistic appropriate in cases where a Wilcoxon-Mann-Whitney test might be used. Here, the pairwise approach would be used in cases where a Kruskal-Wallis test might be used. VDA ranges from 0 to 1, with 0.5 indicating stochastic equality, and 1 indicating that the first group dominates the second. CD ranges from -1 to 1, with 0 indicating stochastic equality, and 1 indicating that the first group dominates the second. rg ranges from -1 to 1, depending on sample size, with 0 indicating no effect, and a positive result indicating that values in the first group are greater than in the second.

Be cautious with this interpretation, as R will alphabetize groups in the formula interface if the grouping variable is not already a factor.

In the function output, VDA.m is the greater of VDA or 1-VDA. CD.m is the absolute value of CD. rg.m is the absolute value of rg.

The function calculates VDA and Cliff's delta from the "W" U statistic from the wilcox.test function. Specifically, VDA = U/(n1*n2); CD = (VDA-0.5)*2.

rg is calculated as 2 times the difference of mean of ranks for each group divided by the total sample size. It appears that rg is equivalent to Cliff's delta.

The input should include either formula and data; or var, and group.

Currently, the function makes no provisions for NA values in the data. It is recommended that NAs be removed beforehand.

Value

A list containing a data frame of pairwise statistics, and the comparison with the most extreme value of the chosen statistic.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The first variable on the right side is used for the grouping variable.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_08.html

See Also

vda, cliffDelta

Examples

data(PoohPiglet)
multiVDA(Likert ~ Speaker, data=PoohPiglet)

Description

Produces McFadden, Cox and Snell, and Nagelkerke pseudo r-squared measures, along with p-values, for models.

Usage

nagelkerke(fit, null = NULL, restrictNobs = FALSE)

Arguments

Details

Pseudo R-squared values are not directly comparable to the R-squared for OLS models. Nor can they be interpreted as the proportion of the variability in the dependent variable that is explained by model. Instead pseudo R-squared measures are relative measures among similar models indicating how well the model explains the data.

Cox and Snell is also referred to as ML. Nagelkerke is also referred to as Cragg and Uhler.

Model objects accepted are lm, glm, gls, lme, lmer, lmerTest, nls, clm, clmm, vglm, glmer, glmmTMB, negbin, zeroinfl, betareg, and rq.

Model objects that require the null model to be defined are nls, lmer, glmer, and clmm. Other objects use the update function to define the null model.

Likelihoods are found using ML (REML = FALSE).

The fitted model and the null model should be properly nested. That is, the terms of one need to be a subset of the the other, and they should have the same set of observations. One issue arises when there are NA values in one variable but not another, and observations with NA are removed in the model fitting. The result may be fitted and null models with different sets of observations. Setting restrictNobs to TRUE ensures that only observations in the fit model are used in the null model. This appears to work for lm and some glm models, but causes the function to fail for other model object types.

Some pseudo R-squared measures may not be appropriate or useful for some model types.

Calculations are based on log likelihood values for models. Results may be different than those based on deviance.

Value

A list of six objects describing the models used, the pseudo r-squared values, the likelihood ratio test for the model, the number of observations for the models, messages, and any warnings.

Acknowledgments

My thanks to Jan-Herman Kuiper of Keele University for suggesting the restrictNobs fix.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/G_10.html

See Also

efronRSquared

Examples

### Logistic regression example
data(AndersonBias)
model = glm(Result ~ County + Gender + County:Gender,
           weight = Count,
           data = AndersonBias,
           family = binomial(link="logit"))
nagelkerke(model)

### Quadratic plateau example 
### With nls, the  null needs to be defined
data(BrendonSmall)
quadplat = function(x, a, b, clx) {
          ifelse(x  < clx, a + b * x   + (-0.5*b/clx) * x   * x,
                           a + b * clx + (-0.5*b/clx) * clx * clx)}
model = nls(Sodium ~ quadplat(Calories, a, b, clx),
            data = BrendonSmall,
            start = list(a   = 519,
                         b   = 0.359,
                         clx = 2304))
nullfunct = function(x, m){m}
null.model = nls(Sodium ~ nullfunct(Calories, m),
             data = BrendonSmall,
             start = list(m   = 1346))
nagelkerke(model, null=null.model)

Description

Defunct. Produces McFadden, Cox and Snell, and Nagelkerke pseudo R-squared measures, along with p-value for the model, for hermite regression objects.

Usage

nagelkerkeHermite(...)

Arguments

Description

Conducts an omnibus symmetry test for a paired contingency table and then post-hoc pairwise tests. This is similar to McNemar and McNemar-Bowker tests in use.

Usage

nominalSymmetryTest(x, method = "fdr", digits = 3, exact = FALSE, ...)

Arguments

Details

The omnibus McNemar test may fail when there are zeros in critical cells.

Currently, the exact=TRUE with a table greater than 2 x 2 will not produce an omnibus test result.

Value

A list containing: a data frame of results of the global test; a data frame of results of the pairwise results; and a data frame mentioning the p-value adjustment method.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/H_05.html

See Also

pairwiseMcnemar, groupwiseCMH, pairwiseNominalIndependence, pairwiseNominalMatrix

Examples

### 2 x 2 repeated matrix example
data(AndersonRainBarrel)
nominalSymmetryTest(AndersonRainBarrel)
                    
### 3 x 3 repeated matrix example
data(AndersonRainGarden)
nominalSymmetryTest(AndersonRainGarden,
                    exact = FALSE)

Description

A data frame with two variables: size of plant nursery in hectares, and proportion of good practices followed by the nursery

Usage

Nurseries

Format

An object of class data.frame with 38 rows and 2 columns.

Source

Mangiafico, S.S., Newman, J.P., Mochizuki, M.J., and Zurawski, D. (2008). Adoption of sustainable practices to protect and conserve water resources in container nurseries with greenhouse facilities. Acta horticulturae 797, 367-372.

Description

Calculates a dominance effect size statistic compared with a theoretical median for one-sample data with confidence intervals by bootstrap

Usage

oneSampleDominance(
  x,
  mu = 0,
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 3,
  na.rm = TRUE,
  ...
)

Arguments

Details

The calculated Dominance statistic is simply the proportion of observations greater than mu minus the the proportion of observations less than mu.

It will range from -1 to 1, with 0 indicating that the median is equal to mu, and 1 indicating that the observations are all greater in value than mu, and -1 indicating that the observations are all less in value than mu.

This statistic is appropriate for truly ordinal data, and could be considered an effect size statistic for a one-sample sign test.

Ordered category data need to re-coded as numeric, e.g. as with as.numeric(Ordinal.variable).

When the statistic is close to 1 or close to -1, or with small sample size, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

VDA is the analogous statistic, converted to a probability, ranging from 0 to 1, specifically, VDA = Dominance / 2 + 0.5.

Value

A small data frame consisting of descriptive statistics, the dominance statistic, and potentially the lower and upper confidence limits.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_03.html

See Also

pairedSampleDominance, cliffDelta, vda

Examples

data(Catbus)
library(DescTools)
SignTest(Catbus$Rating, mu=5.5)
oneSampleDominance(Catbus$Rating, mu=5.5)

Description

Calculates eta-squared as an effect size statistic, following a Kruskal-Wallis test, or for a table with one ordinal variable and one nominal variable; confidence intervals by bootstrap.

Usage

ordinalEtaSquared(
  x,
  g = NULL,
  group = "row",
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 3,
  reportIncomplete = FALSE,
  ...
)

Arguments

Details

Eta-squared is used as a measure of association for the Kruskal-Wallis test or for a two-way table with one ordinal and one nominal variable.

Currently, the function makes no provisions for NA values in the data. It is recommended that NAs be removed beforehand.

eta-squared is typically positive, though may be negative in some cases, as is the case with adjusted r-squared. It's not recommended that the confidence interval be used for statistical inference.

When eta-squared is close to 0 or very large, or with small counts in some cells, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

Value

A single statistic, eta-squared. Or a small data frame consisting of eta-squared, and the lower and upper confidence limits.

Note

Note that eta-squared as calculated by this function is equivalent to the epsilon-squared, or adjusted r-squared, as determined by an anova on the rank-transformed values. Eta-squared for Kruskal-Wallis is typically defined this way in the literature.

Author(s)

Salvatore Mangiafico, [email protected]

References

Cohen, B.H. 2013. Explaining Psychological Statistics, 4th ed. Wiley.

https://rcompanion.org/handbook/F_08.html

See Also

freemanTheta, epsilonSquared

Examples

data(Breakfast)
library(coin)
chisq_test(Breakfast, scores = list("Breakfast" = c(-2, -1, 0, 1, 2)))
ordinalEtaSquared(Breakfast)

data(PoohPiglet)
kruskal.test(Likert ~ Speaker, data = PoohPiglet)
ordinalEtaSquared(x = PoohPiglet$Likert, g = PoohPiglet$Speaker)

### Same data, as matrix of counts
data(PoohPiglet)
XT = xtabs( ~ Speaker + Likert , data = PoohPiglet)
ordinalEtaSquared(XT)

Description

Calculates a dominance effect size statistic for two-sample paired data with confidence intervals by bootstrap

Usage

pairedSampleDominance(
  formula = NULL,
  data = NULL,
  x = NULL,
  y = NULL,
  ci = FALSE,
  conf = 0.95,
  type = "perc",
  R = 1000,
  histogram = FALSE,
  digits = 3,
  na.rm = TRUE,
  ...
)

Arguments

Details

The calculated Dominance statistic is simply the proportion of observations in x greater the paired observations in y, minus the proportion of observations in x less than the paired observations in y

It will range from -1 to 1, with and 1 indicating that the all the observations in x are greater than the paired observations in y, and -1 indicating that the all the observations in y are greater than the paired observations in x.

The input should include either formula and data; or x, and y. If there are more than two groups, only the first two groups are used.

This statistic is appropriate for truly ordinal data, and could be considered an effect size statistic for a two-sample paired sign test.

Ordered category data need to re-coded as numeric, e.g. as with as.numeric(Ordinal.variable).

When the statistic is close to 1 or close to -1, or with small sample size, the confidence intervals determined by this method may not be reliable, or the procedure may fail.

VDA is the analogous statistic, converted to a probability, ranging from 0 to 1, specifically, VDA = Dominance / 2 + 0.5

Value

A small data frame consisting of descriptive statistics, the dominance statistic, and potentially the lower and upper confidence limits.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_07.html

See Also

oneSampleDominance, vda, cliffDelta

Examples

data(Pooh)
Time.1 = Pooh$Likert[Pooh$Time == 1]
Time.2 = Pooh$Likert[Pooh$Time == 2]
library(DescTools)
SignTest(x = Time.1, y = Time.2)
pairedSampleDominance(x = Time.1, y = Time.2)
pairedSampleDominance(Likert ~ Time, data=Pooh)

Description

Defunct. Calculates the differences in the response variable for each pair of levels of a grouping variable in an unreplicated complete block design.

Usage

pairwiseDifferences(...)

Arguments

Description

Conducts pairwise McNemar, exact, and permutation tests as a post-hoc to Cochran Q test.

Usage

pairwiseMcnemar(
  formula = NULL,
  data = NULL,
  x = NULL,
  g = NULL,
  block = NULL,
  test = "exact",
  method = "fdr",
  digits = 3,
  correct = FALSE
)

Arguments

Details

The component tables for the pairwise tests must be of size 2 x 2.

The input should include either formula and data; or x, g, and block.

Value

A list containing: a data frame of results of the global test; a data frame of results of the pairwise results; and a data frame mentioning the p-value adjustment method.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The first variable on the right side is used for the grouping variable. The second variable on the right side is used for the blocking variable.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/H_07.html

See Also

nominalSymmetryTest, groupwiseCMH, pairwiseNominalIndependence, pairwiseNominalMatrix

Examples

### Cochran Q post-hoc example
data(HayleySmith)
library(DescTools)
CochranQTest(Response ~ Practice | Student,
             data = HayleySmith)
HayleySmith$Practice = factor(HayleySmith$Practice,
                          levels = c("MowHeight", "SoilTest",
                                     "Clippings", "Irrigation"))
PT = pairwiseMcnemar(Response ~ Practice | Student,
                     data    = HayleySmith,
                     test    = "exact",
                     method  = "fdr",
                     digits  = 3)
PT
PT = PT$Pairwise
cldList(comparison = PT$Comparison,
        p.value    = PT$p.adjust,
        threshold  = 0.05)

Description

Conducts pairwise Mood's median tests across groups.

Usage

pairwiseMedianMatrix(
  formula = NULL,
  data = NULL,
  x = NULL,
  g = NULL,
  digits = 4,
  method = "fdr",
  ...
)

Arguments

Details

The input should include either formula and data; or x, and g.

Mood's median test compares medians among two or more groups. See https://rcompanion.org/handbook/F_09.html for futher discussion of this test.

The pairwiseMedianMatrix function can be used as a post-hoc method following an omnibus Mood's median test. It passes the data for pairwise groups to coin::median_test.

The matrix output can be converted to a compact letter display, as in the example.

Value

A list consisting of: a matrix of p-values; the p-value adjustment method; a matrix of adjusted p-values.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The first variable on the right side is used for the grouping variable.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_09.html

See Also

pairwiseMedianTest

Examples

data(PoohPiglet)
PoohPiglet$Speaker = factor(PoohPiglet$Speaker,
                          levels = c("Pooh", "Tigger", "Piglet"))
PT = pairwiseMedianMatrix(Likert ~ Speaker,
                          data   = PoohPiglet,
                          exact  = NULL,
                          method = "fdr")$Adjusted
PT                           
library(multcompView)
multcompLetters(PT,
                compare="<",
                threshold=0.05,
                Letters=letters)

Description

Conducts pairwise Mood's median tests across groups.

Usage

pairwiseMedianTest(
  formula = NULL,
  data = NULL,
  x = NULL,
  g = NULL,
  digits = 4,
  method = "fdr",
  ...
)

Arguments

Details

The input should include either formula and data; or x, and g.

Mood's median test compares medians among two or more groups. See https://rcompanion.org/handbook/F_09.html for further discussion of this test.

The pairwiseMedianTest function can be used as a post-hoc method following an omnibus Mood's median test. It passes the data for pairwise groups to coin::median_test.

The output can be converted to a compact letter display, as in the example.

Value

A dataframe of the groups being compared, the p-values, and the adjusted p-values.

Note

The parsing of the formula is simplistic. The first variable on the left side is used as the measurement variable. The first variable on the right side is used for the grouping variable.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/F_09.html

See Also

pairwiseMedianMatrix

Examples

data(PoohPiglet)
PoohPiglet$Speaker = factor(PoohPiglet$Speaker,
                     levels = c("Pooh", "Tigger", "Piglet"))
PT = pairwiseMedianTest(Likert ~ Speaker,
                        data   = PoohPiglet,
                        exact  = NULL,
                        method = "fdr")
PT                         
cldList(comparison = PT$Comparison,
        p.value    = PT$p.adjust,
        threshold  = 0.05)

Description

Compares a series of models with pairwise F tests and likelihood ratio tests.

Usage

pairwiseModelAnova(fits, ...)

Arguments

Details

For comparisons to be valid, both models must have the same data, without transformations, use the same dependent variable, and be fit with the same method.

To be valid, models need to be nested.

Value

A list of: The calls of the models compared; a data frame of comparisons and F tests; and a data frame of comparisons and likelihood ratio tests.

Author(s)

Salvatore Mangiafico, [email protected]

See Also

compareGLM, compareLM

Examples

### Compare among polynomial models
data(BrendonSmall)
BrendonSmall$Calories = as.numeric(BrendonSmall$Calories)

BrendonSmall$Calories2 = BrendonSmall$Calories * BrendonSmall$Calories
BrendonSmall$Calories3 = BrendonSmall$Calories * BrendonSmall$Calories * 
                         BrendonSmall$Calories
BrendonSmall$Calories4 = BrendonSmall$Calories * BrendonSmall$Calories * 
                         BrendonSmall$Calories * BrendonSmall$Calories
model.1 = lm(Sodium ~ Calories, data = BrendonSmall)
model.2 = lm(Sodium ~ Calories + Calories2, data = BrendonSmall)
model.3 = lm(Sodium ~ Calories + Calories2 + Calories3, data = BrendonSmall)
model.4 = lm(Sodium ~ Calories + Calories2 + Calories3 + Calories4,
             data = BrendonSmall)
pairwiseModelAnova(model.1, model.2, model.3, model.4)

Description

Conducts pairwise tests for a 2-dimensional matrix, in which at at least one dimension has more than two levels, as a post-hoc test. Conducts Fisher exact, Chi-square, or G-test.

Usage

pairwiseNominalIndependence(
  x,
  compare = "row",
  fisher = TRUE,
  gtest = TRUE,
  chisq = TRUE,
  method = "fdr",
  correct = "none",
  yates = FALSE,
  stats = FALSE,
  cramer = FALSE,
  digits = 3,
  ...
)

Arguments

Value

A data frame of comparisons, p-values, and adjusted p-values.

Acknowledgments

My thanks to Carole Elliott of Kings Park & Botanic Gardens for suggesting the inclusion on the chi-square statistic and degrees of freedom in the output.

Author(s)

Salvatore Mangiafico, [email protected]

References

https://rcompanion.org/handbook/H_04.html

See Also

pairwiseMcnemar, groupwiseCMH, nominalSymmetryTest, pairwiseNominalMatrix

Examples

### Independence test for a 4 x 2 matrix
data(Anderson)
fisher.test(Anderson)
Anderson = Anderson[(c("Heimlich", "Bloom", "Dougal", "Cobblestone")),]
PT = pairwiseNominalIndependence(Anderson,
                                 fisher = TRUE,
                                 gtest  = FALSE,
                                 chisq  = FALSE,
                                 cramer = TRUE)
PT                                
cldList(comparison = PT$Comparison,
        p.value    = PT$p.adj.Fisher,
        threshold  = 0.05)

Description

Conducts pairwise tests for a 2-dimensional matrix, in which at at least one dimension has more than two levels, as a post-hoc test. Conducts Fisher exact, Chi-square, or G-test.

Usage

pairwiseNominalMatrix(
  x,
  compare = "row",
  fisher = TRUE,
  gtest = FALSE,
  chisq = FALSE,
  method = "fdr",
  correct = "none",
  digits = 3,
  ...
)