Question 1

What is the most common data format encountered in statistics with R, from among the following?
A) .dta
B) .sas7bdat
C) .csv
D) .sav

Accepted Answer

The comma-separated values format allows easy transfer between R and various other statistical packages, spreadsheets, and other applications.

Question 2

What could be an "object" in R?
A) The output from a statistical procedure
B) A vector of values of a variable
C) A data frame
D) Any of the above

Accepted Answer

An object is a data structure with (1) attributes and (2) suitability for methods which act on its attributes. All objects are of some "class". View this by typing class(name_of_object). A class may be thought of as a type of "blueprint" for an object. Any given statistical method will only work on certain classes of objects for which it was designed.

Question 3

If race is a numerically coded variable, what does the command racef <- factor(race) do?
A) Recodes race so 0 is the first value
B) Converts race into a categorical variable
C) Prints the level labels for race
D) Prints the codebook for race

Accepted Answer

is the correct answer. Factors are categorical variables.
racef <- as.factor(survey$race)
class(survey$race)
[1] "integer"
class(racef)
[1] "factor"

Question 4

"Fitted values" are…
A) observed values
B) standardized values
C) predicted values
D) centered values

Accepted Answer

Observed values are raw values and are the basis for prediction. Standardized values are a transformation of raw values implemented by subtracting the mean and dividing by the standard deviation, resulting in variables with means of 0 and standard deviations of 1. Centered values are similar but only subtract the mean and do not divide by the standard deviation, resulting in variables with means of 0 but different variances.

Question 5

What command in R is used to implement a binary logistic regression model?
A) logistic
B) blogistic
C) binomial
D) glm

Accepted Answer

The glm command in R implements generalized linear models (GZLM), of which logistic regression is one type. While GZLM does include OLS regression it is mainly used for various types of nonlinear regression. Logistic regression is nonlinear in the sense that raw values are not linear with the untransformed values of the DV. In GZLM the link function transforms the DV. For logistic regression, the log odds of the raw DV is made the outcome variable. The goal of transformation is to make the raw values of the predictors linear with the transformed DV. This explains why the "L" in GZLM is "linear" yet we may say GZLM supports certain types of nonlinear regression.
The logistic command was not discussed in the text but is from the "psych" package and does not implement logistic regression. Rather it returns probabilities of a value (x) based on the logistic distribution. This command is used in item response theory (IRT) models. The "stats" package in R's system library has a "Logistic" family of commands (dlogis, plogis, qlogis, and rlogis) which return various forms of the logistic distribution but do not implement logistic regression.
There is no blogistic command, though there is a blogistic blog on the internet. Likewise, there is no binomial command to implement logistic regression. When used as in binomial(link = "logit"), the binomial function is specifying a distribution family. Also, the "stats" package in R's system library has a "Binomial" family of commands (dbinom, pbinom, qbinom, and rbinom) which return various forms of the binomial distribution but do not implement logistic regression.

Question 6

In logistic regression and elsewhere, deviance is an important measure of model fit (and model error). Which is not synonymous with deviance?
A) the regressor
B) -2LL
C) model chi-square
D) likelihood ratio

Accepted Answer

All the others are synonyms for deviance.

Question 7

What does the relevel command do?
A) Sets the reference level for a categorical variable
B) Standardizes a variable
C) Centers a variable
D) Converts a variable to a flat distribution

Accepted Answer

In many forms of regression for categorical variables a reference level must be declared. Coefficients for other levels of a categorical variable are then interpreted in relation to the reference level. For instance, consider predicting income (a continuous variable) from race (1 = White, 2 = African-American, 3 = Other), where the "1" category is the reference level for race, then a coefficient of -0.7 for African-American would mean that compared to the reference level (White), African-Americans are predicted to have lower income.

Question 8

In a well-fitting model what kind of distribution should this R command display?: hist(olsmodel$residuals)
A) Dots around a 45-degree line
B) A bi-modal distribution
C) A normal distribution
D) A random distribution

Accepted Answer

In any well-fitting predictive model, residuals should be normally distributed. This is because to be well-fitting, most predictions should be close to observed values and residuals (observed minus expected) should be close to zero. High and low predictions should trail off in either direction, forming the bell-shaped curve which characterizes a normal distribution.

Question 9

To compute an estimated marginal mean, sum the observed values and divide by n.

Accepted Answer

The statement would return the raw mean. Estimated marginal means are means controlling for other variables in the model.

Question 10

For what type of data is gamma regression used?
A) heteroskedastic
B) skewed
C) missing
D) count

Accepted Answer

Poisson or negative binomial regression is used for count data. Answers A and C are problems for regression models in general, not particular to gamma regression.

Question 11

What model is used for overdispersed count data?
A) gamma
B) poisson
C) negative binomial
D) cloglog

Accepted Answer

By rule of thumb, overdispersion occurs when the variance of the DV is greater than its mean. In statistical packages, overdispersion is calculated using the deviance. When count data are overdispersed, Poisson regression is dropped in favor of some other technique, usually negative binomial regression.

Question 12

What R command mainly used in the textbook to implements multilevel (mixed linear) models?
A) lm
B) lmer
C) hsb
D) mixed

Accepted Answer

though there are other R commands which also implement multilevel models. The lm command is for ordinary (monolevel) linear regression. There is no hsb( ) command but in the "merTools" package, which was not discussed in the text, the command data(hsb) will invoke the High School and Beyond dataset, which was used in the text. There is a mixed command in the "afex" package, but this was not discussed in the text. That command implements variations on multilevel modeling using the lmer command.

Question 13

What R command is mainly used in the textbook to implement panel data regression models?
A) plm
B) lmer
C) lm
D) glm

Accepted Answer

B is for multilevel modeling. C is for OLS regression. D is for GZLM models.

Question 14

In panel data regression using the leading R package for this procedure, when asking for a fixed effects model of subject-level effects, what are the model= and effect= specifications?
A) model="within", effect="individual"
B) model="within", effect="year"
C) model="between", effect="individual"
D) model="between", effect="year"

Accepted Answer

The question refers to the plm command.

Question 15

In panel data regression using the leading R package for this procedure, what do idiosyncratic effects refer to?
A) random effects variance at the individual level
B) fixed effects variance at the individual level
C) random effects variance not explained by the subject (individual) variable or fixed effects
D) fixed effects, as opposed to variance components

Accepted Answer

The question refers to the plm command.

Question 16

In a structural model with standardized structural (path) coefficients, what is the value of a compound (indirect) path?
A) the sum of the structural coefficients
B) the product of the structural coefficients
C) 1.0
D) This value is always indeterminate.

Accepted Answer

This is called the "multiplication rule",

Question 17

In the text supplement for Chapter 3, which R command is primarily used to implement structural equation models?
A) structural
B) structure
C) lavaan
D) path

Accepted Answer

The lavaan command from the package "lavaan" has been the primary SEM package in R and is the one emphasized in the textbook. A is not an R command. B is a command from R's "base" package and returns the attributes of an object. D is also wrong. While SEM may implement path analysis, the path command, not discussed in the text, is from the "igraph" package and is a helper function to add or delete edges along a path in a graphic or network model. There are other packages to fit structural equation models in R, notably the sem command from the "sem" package. There is also a sem command in the lavaan package itself, and there are a large number of other SEM-related packages and commands in R.

Quiz 3: Statistical Analytics With R Part 2

What is the most common data format encountered in statistics with R, from among the following?

What could be an "object" in R?

If race is a numerically coded variable, what does the command racef <- factor(race) do?

"Fitted values" are…

What command in R is used to implement a binary logistic regression model?

In logistic regression and elsewhere, deviance is an important measure of model fit (and model error) . Which is not synonymous with deviance?

What does the relevel command do?

In a well-fitting model what kind of distribution should this R command display?: hist(olsmodel$residuals)

To compute an estimated marginal mean, sum the observed values and divide by n.

For what type of data is gamma regression used?

What model is used for overdispersed count data?

What R command mainly used in the textbook to implements multilevel (mixed linear) models?

What R command is mainly used in the textbook to implement panel data regression models?

In panel data regression using the leading R package for this procedure, when asking for a fixed effects model of subject-level effects, what are the model= and effect= specifications?

In panel data regression using the leading R package for this procedure, what do idiosyncratic effects refer to?

In a structural model with standardized structural (path) coefficients, what is the value of a compound (indirect) path?

In the text supplement for Chapter 3, which R command is primarily used to implement structural equation models?