Datasets from Hosmer, Lemeshow and Sturdivant, "Applied Logistic Regression" (3rd ed., 2013)

Description

This package is an unofficial companion to the textbook "Applied Logistic Regression" by D.W. Hosmer, S. Lemeshow and R.X. Sturdivant (3rd ed., 2013).

Details

It includes all the datasets used in the book, both for easy reproducibility and algorithms benchmarking purposes.

Some analysis proposed in the text are reproduced in the examples, in order to provide data testing and code demos at the same time.

The vignette includes all the examples (with graphics too); therefore is organized per-dataset.

Datasets and variables have lower-case name with respect to the original sources. Categorical data were packaged as factor.

Regarding data coding, help pages list the internal/factor representation of the data (eg 1: No, 2: Yes), not the original one (eg 0: No, 1: Yes). This is intended to allow easier/safer recoding based on as.integer, especially for multinomial variables.

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

APS data

Description

aps dataset.

Usage

aps

Format

A data.frame with 508 rows and 11 variables:

id

Identification Code (1 - 508)

place

Placement (1: Outpatient, 2: Day Treatment, 3: Intermediate Residential, 4: Residential)

place3

Placement Combined (1: Outpatient or Day Treatment, 2: Intermediate Residential, 3: Residential )

age

Age at Admission (Years)

race

Race (1: White, 2: Non-white)

gender

Gender (1: Female, 2: Male)

neuro

Neuropsychiatric Disturbance (1: None, 2: Mild, 3: Moderate, 4: Severe)

emot

Emotional Disturbance (1: Not Severe, 2: Severe)

danger

Danger to Others (1: Unlikely, 2: Possible, 3: Probable, 4: Likely)

elope

Elopement Risk (1: No Risk, 2: At Risk)

los

Length of Hospitalization (Days)

behav

Behavioral Symptoms Score (0 - 9)

custd

State Custody (1: No, 2: Yes)

viol

History of Violence (1: No, 2: Yes)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(aps, n = 10)
summary(aps)


## Table 8.2 p. 274
library(nnet)
modt8.2 <- multinom(place3 ~ viol, data = aps)
summary(modt8.2)
exp(coef(modt8.2)[, "violYes"])
t(exp(confint(modt8.2)["violYes", ,]))
## To test differences between b_2 and b_1 we need the estimated variance
## covariance matrix for the fitted model (Table 8.3 p. 274). 
vcov(modt8.2) # 'raw'
## To have exactly the same output as the text we need to rearrange just a
## minimum
VarCovM <- vcov(modt8.2)[c(2, 1, 4, 3), c(2, 1, 4, 3)]
VarCovM[upper.tri(VarCovM)] <- NA
VarCovM
## Testing against null model. 
modt8.2Null <- multinom(place3 ~ 1, data = aps)
anova(modt8.2, modt8.2Null, test = "Chisq")

BURN1000 data

Description

burn1000 dataset.

Usage

burn1000

Format

A data.frame with 1000 rows and 9 variables:

id

Identification code (1 - 1000)

facility

Burn facility (1 - 40)

death

Hospital discharge status (1: Alive, 2: Dead)

age

Age at admission (Years)

gender

Gender (1: Female, 2: Male)

race

Race (1: Non-White, 2: White)

tbsa

Total burn surface area (0 - 100%)

inh_inj

Burn involved inhalation injury (1: No, 2: Yes)

flame

Flame involved in burn injury (1: No, 2: Yes)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(burn1000, n = 10)
summary(burn1000)

## Table 3.15 p. 80
summary(mod3.15 <- glm(death ~ tbsa + inh_inj + age + gender + flame + race,
                       family = binomial, data = burn1000 ))

BURN13M data

Description

burn13m dataset.

Usage

burn13m

Format

A data.frame with 388 rows and 11 variables: the covariate are the same as those from burn1000 with the addition of

pair

Pair Identification Code (1-119)

pairid

Subject Identification Code within pair (1-4)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(burn13m, n = 10)
summary(burn13m)

BURN_EVAL_1 data

Description

burn_eval_1 dataset.

Usage

burn_eval_1

Format

A data.frame with 500 rows and 9 variables: the covariate are the same as those from burn1000.

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(burn_eval_1, n = 10)
summary(burn_eval_1)

BURN_EVAL_2 data

Description

burn_eval_2 dataset.

Usage

burn_eval_2

Format

A data.frame with 500 rows and 9 variables: the covariate are the same as those from burn1000.

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(burn_eval_2, n = 10)
summary(burn_eval_2)

CHDAGE data

Description

chdage dataset.

Usage

chdage

Format

A data.frame with 100 rows and 4 variables:

id

Identification code (1 - 100)

age

Age (Years)

agegrp

Age group (1: 20-39, 2: 30-34, 3: 35-39, 4: 40-44, 5: 45-49, 6: 50-54, 7: 55-59, 8: 60-69)

chd

Presence of CHD (1: No, 2: Yes)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(chdage,  n = 10)
summary(chdage)

## Figure 1.1 p. 5
plot(as.integer(chd)-1 ~ age,
     pch = 20,
     main = "Figure 1.1 p. 5",
     ylab = "Coronary heart disease",
     xlab = "Age (years)",
     data = chdage)

## Table 1.2
with(chdage, addmargins(table(agegrp)))
with(chdage, addmargins(table(agegrp, chd)))
(Means <- with(chdage, tapply(as.integer(chd)-1, list(agegrp), mean)))

## Figure 1.2 p. 6
midPoints <- c(24.5, seq(32, 57, 5), 64.5)
plot(midPoints, Means, pch = 20,
     ylab = "Coronary heart disease (mean)",
     xlab = "Age (years)", ylim = 0:1,
     main = "Figure 1.2 p. 6")
lines(midPoints, Means)

## Table 1.3
summary( mod1.3 <- glm( chd ~ age, family = binomial, data = chdage ))

## Table 1.4
vcov(mod1.3)

## Computing OddsRatio and confidence intervals for age ...
exp(coef(mod1.3))[-1]
exp(confint(mod1.3))[-1, ]

GLOW11M data

Description

glow11m dataset.

Usage

glow11m

Format

A data.frame with 238 rows and 16 variables: the covariate are the same as those from glow500 with the addition of

pair

Pair Identification Code (1-119)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(glow11m, n = 10)
summary(glow11m)

## Table 7.2 p. 252
library(survival)
mod7.2 <- clogit(as.numeric(fracture) ~ height + weight + bmi +
                 priorfrac + premeno + momfrac + armassist + raterisk +
                 strata(pair), data = glow11m)
summary(mod7.2)

GLOW500 data

Description

glow500 dataset.

Usage

glow500

Format

A data.frame with 500 rows and 15 variables:

sub_id

Identification Code (1 - n)

site_id

Study Site (1 - 6)

phy_id

Physician ID code (128 unique codes)

priorfrac

History of Prior Fracture (1: No, 2: Yes)

age

Age at Enrollment (Years)

weight

Weight at enrollment (Kilograms)

height

Height at enrollment (Centimeters)

bmi

Body Mass Index (Kg/m^2)

premeno

Menopause before age 45 (1: No, 2: Yes)

momfrac

Mother had hip fracture (1: No, 2: Yes)

armassist

Arms are needed to stand from a chair (1: No, 2: Yes)

smoke

Former or current smoker (1: No, 2: Yes)

raterisk

Self-reported risk of fracture (1: Less than others of the same age, 2: Same as others of the same age, 3: Greater than others of the same age)

fracscore

Fracture Risk Score (Composite Risk Score)

fracture

Any fracture in first year (1: No, 2: Yes)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(glow500, n = 10)
summary(glow500)

## Table 2.2 p. 39
summary(mod2.2 <- glm(fracture ~ age + weight + priorfrac +
                                 premeno + raterisk,
                      family = binomial,
                      data = glow500))

## Table 2.3 p. 40
summary(mod2.3 <- update(mod2.2, . ~ . - weight - premeno))

## Table 2.4 p. 44
vcov(mod2.3)

## Table 3.6 p. 58
contrasts(glow500$raterisk)

## Contrasts: Table 3.8 and 3.9 p. 60
contrasts(glow500$raterisk) <- matrix(c(-1,-1,1,0,0,1), byrow= TRUE, ncol = 2)
summary(mod3.9 <- glm(fracture ~ raterisk, family = binomial,
                      data = glow500))
# cleaning modified dataset ...
rm(glow500)

## Table 5.1 pg 160 - Hosmer-Lemeshow test (with vcdExtra package)
mod4.16 <- glm(fracture ~ age * priorfrac + height + momfrac * armassist +
                          I(as.integer(raterisk) == 3) ,
               family = binomial,
               data = glow500)
library(vcdExtra)
summary(HLtest(mod4.16))

## Table 5.3 p. 171 - Classification table
glow500$pred4.16 <- predict(mod4.16, type = "response")
with(glow500, addmargins(table( pred4.16 > 0.5, fracture)))

## Sensitivy, specificity, ROC (using pROC)
library(pROC)

## Figure 5.3 p. 177 - ROC curve (using pROC package)
print(roc4.16 <- roc(fracture ~ pred4.16, data = glow500))
plot(roc4.16, main = "Figure 5.3 p. 177")

## Table 5.8 p. 175
vars <- c("thresholds","sensitivities","specificities")
tab5.8 <- data.frame(roc4.16[vars])
## Now, for printing/comparison purposes, steps below in order to find
## threshold values most similar to those in the table
findIndex <- function(x, y) which.min( (x-y)^2 )
cutPoints <- seq(0.05, 0.75, by = 0.05)
tableIndex <- mapply(findIndex, y = cutPoints,
                     MoreArgs = list(x = roc4.16$thresholds))
## And finally, let's print a reasonable approximation of table 5.8
writeLines("\nTable 5.8 p. 175\n")
tab5.8[tableIndex, ]

## Figure 5.1 p. 175
plot(specificities ~ thresholds, xlim = c(0, 1), type = "l",
     xlab = "Probabilty cutoff", ylab = "Sensitivity/specificity",
     ylim = c(0, 1), data = tab5.8, main = "Figure 5.1 p. 175")
with(tab5.8, lines(thresholds, sensitivities, col = "red"))
legend(x = 0.75, y = 0.55, legend = c("Sensitivity", "Specificity"),
       lty = 1, col = c("red","black"))
abline(h = c(0, 1), col = "grey80", lty = "dotted")

GLOW_BONEMED data

Description

glow_bonemed dataset.

Usage

glow_bonemed

Format

A data.frame with 500 rows and 18 variables: the covariate are the same as those from glow500 with the addition of

bonemed

Bone medications at enrollment (1: No, 2: Yes)

bonemed_fu

Bone medications at follow-up (1: No, 2: Yes)

bonetreat

Bone medications both at enrollment and follow-up (1: No, 2: Yes)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(glow_bonemed, n = 10)
summary(glow_bonemed)

GLOW_MIS_COMP data

Description

glow_mis_comp dataset.

Usage

glow_mis_comp

Format

A data.frame with 500 rows and 10 variables: the covariate are the same as those from glow500, without bmi, premeno, armassist, smoke and fracscore.

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(glow_mis_comp, n = 10)
summary(glow_mis_comp)

GLOW_MIS_WMISSING data

Description

glow_mis_wmissing dataset.

Usage

glow_mis_wmissing

Format

A data.frame with 500 rows and 10 variables: the covariate are the same as those from glow500, without bmi, premeno, armassist, smoke and fracscore.

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(glow_mis_wmissing, n = 10)
summary(glow_mis_wmissing)

GLOW_RAND data

Description

glow_rand dataset.

Usage

glow_rand

Format

A data.frame with 500 rows and 15 variables: the covariate are the same as those from glow500.

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(glow_rand, n = 10)
summary(glow_rand)

ICU data

Description

icu dataset.

Usage

icu

Format

A data.frame with 200 rows and 21 variables:

id

Identification code (ID Number)

sta

Vital Status at hospital discharge (1: Lived, 2: Died)

age

Age (Years)

gender

Gender (1: Male, 2: Female)

race

Race (1: White, 2: Black, 3: Other)

ser

Service at ICU admission (1: Medical, 2: Surgical)

can

Cancer part of present problem (1: No, 2: Yes)

crn

History of chronic renal failure (1: No, 2: Yes)

inf

Infection probable at ICU admission (1: No, 2: Yes)

cpr

CPR prior to ICU admission (1: No, 2: Yes)

sys

Systolic blood pressure at ICU admission (mm Hg)

hra

Heart rate at ICU admission (Beats/min)

pre

Previous admission to an ICU within 6 months (1: No, 2: Yes)

type

Type of admission (1: Elective, 2: Emergency)

fra

Long bone, multiple, neck, single area, or hip fracture (1: No, 2: Yes)

po2

PO2 from initial blood gases (1: > 60, 2: <= 60)

ph

PH from initial blood gases (1: >= 7.25, 2: < 7.25)

pco

PCO2 from initial blood gases (1: <= 45, 2: > 45)

bic

Bicarbonate from initial blood gases (1: >= 18, 2: < 18)

cre

Creatinine from initial blood gases (1: <= 2.0, 2: > 2.0)

loc

Level of consciousness at ICU admission (1: No coma or deep stupor, 2: Deep stupor, 3: Coma)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(icu, n = 10)
summary(icu)

LOWBWT data

Description

lowbwt dataset.

Usage

lowbwt

Format

A data.frame with 189 rows and 11 variables:

id

Identification Code

low

Low birth weight (1: >= 2500, 2: < 2500 g)

age

Age of mother (Years)

lwt

Weight of mother at last menstrual period (Pounds)

race

Race (1: White, 2: Black, 3: Other)

smoke

Smoking status during pregnancy (1: No, 2: Yes)

ptl

History of premature labor (1: None, 2: One, 3: Two, etc)

ht

History of hypertension (1: No, 2: Yes)

ui

Presence of Uterine irritability (1: No, 2: Yes)

ftv

Number of physician visits during the first trimester (1: None, 2: One, 3: Two, etc)

bwt

Recorded birth weight (Grams)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(lowbwt, n = 10)
summary(lowbwt)

MYOPIA data

Description

myopia dataset.

Usage

myopia

Format

A data.frame with 618 rows and 18 variables:

id

Subject identifier (1-1503)

studyyear

Year subject entered the study (Year)

myopic

Myopia within the first five years of follow up (1: No, 2: Yes)

age

Age at first visit (Years)

gender

Gender (1: Male, 2: Female)

spheq

Spherical Equivalent Refraction (diopter)

al

Axial Length (mm)

acd

Anterior Chamber Depth (mm)

lt

Lens Thickness (mm)

vcd

Vitreous Chamber Depth (mm)

sporthr

How many hours per week outside of school the child spent engaging in sports/outdoor activities (Hours per week)

readhr

How many hours per week outside of school the child spent reading for pleasure (Hours per week)

comphr

How many hours per week outside of school the child spent playing video/computer games or working on the computer (Hours per week)

studyhr

How many hours per week outside of school the child spent reading or studying for school assignments (Hours per week)

tvhr

How many hours per week outside of school the child spent watching television (Hours per week)

diopterhr

Composite of near-work activities (Hours per week)

mommy

Was the subject's mother myopic? (1: No, 2: Yes)

dadmy

Was the subject's father myopic? (1: No, 2: Yes)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(myopia, n = 10)
summary(myopia)

NHANES data

Description

nhanes dataset.

Usage

nhanes

Format

A data.frame with 6482 rows and 21 variables:

id

Identification Code (1 - 6482)

gender

Gender (1: Male, 2: Female)

age

Age at Screening (Years)

marstat

Marital Status (1: Married, 2: Widowed, 3: Divorced, 4: Separated, 5: Never Married, 6: Living Together)

samplewt

Statistical Weight (4084.478 - 153810.3)

psu

Pseudo-PSU (1, 2)

strata

Pseudo-Stratum (1 - 15)

tchol

Total Cholesterol (mg/dL)

hdl

HDL-Cholesterol (mg/dL)

sysbp

Systolic Blood Pressure (mm Hg)

dbp

Diastolic Blood Pressure (mm Hg)

wt

Weight (kg)

ht

Standing Height (cm)

bmi

Body mass Index (Kg/m^2)

vigwrk

Vigorous Work Activity (1: Yes, 2: No)

modwrk

Moderate Work Activity (1: Yes, 2: No)

wlkbik

Walk or Bicycle (1: Yes, 2: No)

vigrecexr

Vigorous Recreational Activities (1: Yes, 2: No)

modrecexr

Moderate Recreational Activities (1: Yes, 2: No)

sedmin

Minutes of Sedentary Activity per Week (1: Yes, 2: No)

obese

BMI>35 (1: No, 2: Yes)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(nhanes, n = 10)
summary(nhanes)

POLYPHARM data

Description

polypharm dataset.

Usage

polypharm

Format

A data.frame with 3500 rows and 14 variables:

id

Subject ID (1 - 500)

polypharmacy

Outcome; taking drugs from more than three different classes (1: No, 2: Yes)

mhv4

Number of outpatient Mental Health Visits (1: none, 2: one to five, 3: six to fourteen, 4: greater than 14)

inptmhv3

Number of inpatient Mental Health Visits (1: none, 2: one, 3: more than one)

year

Year (2002 to 2008)

group

Group (1: Covered Families and Children - CFC, 2: Aged, Blind or Disabled - ABD, 3: Foster Care - FOS)

urban

Location (1: Urban, 2: Rural)

comorbid

Comorbidity (1: No, 2: Yes)

anyprim

Any primary diagnosis (bipolar, depression, etc.) (1: No, 2: Yes)

numprim

Number of primary diagnosis (1: none, 2: one, 3: more than one)

gender

Gender (1: Female, 2: Male)

race

Race (1: White, 2: Black, 3: Other)

ethnic

Ethnic category (1: Non-Hispanic, 2: Hispanic)

age

Age (Years and months, two decimal places)

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(polypharm, n = 10)
summary(polypharm)

SCALE_EXAMPLE data

Description

scale_example dataset.

Usage

scale_example

Format

A data.frame with 500 rows and 2 variables:

y

a dicotomic variable (say 1: No, 2: Yes)

x

a numeric variable

Source

Hosmer, D.W., Lemeshow, S. and Sturdivant, R.X. (2013) Applied Logistic Regression, 3rd ed., New York: Wiley

Examples

head(scale_example, n = 10)
summary(scale_example)