Type: Package
Title: Turn a Regression Model Inside Out
Version: 1.1.1
Maintainer: David Melamed <dmmelamed@gmail.com>
Description: Turns regression models inside out. Functions decompose variances and coefficients for various regression model types. Functions also visualize regression model objects using techniques developed in Schoon, Melamed, and Breiger (2024) <doi:10.1017/9781108887205>.
VignetteBuilder: knitr
Depends: R (≥ 3.5.0), ggplot2, methods
Suggests: dplyr, knitr, rmarkdown, ggrepel, MASS
License: GPL-2 | GPL-3
Encoding: UTF-8
LazyData: true
NeedsCompilation: no
Packaged: 2024-06-13 15:18:34 UTC; melamed.9
Author: David Melamed ORCID iD [aut, cre], Ronald L. Breiger [aut], Eric W. Schoon [aut]
Repository: CRAN
Date/Publication: 2024-06-14 09:50:01 UTC

Replication data for Beckfield (2006) as re-analyzed by Schoon, Melamed, and Breiger (2024)

Description

Beckfield (2006) analyzed these data using fixed and random effects regression models. He showed that regional economic and political integregation is associated with increased economic inequality. Schoon, Melamed, and Breiger (2024) turned these models inside out and decomposed the model coefficients.

Usage

data("Beckfield06")

Format

A data frame with 48 observations on the following 9 variables.

year

a numeric vector

polint

a numeric vector

ecoint

a numeric vector

ecoints

a numeric vector

gdp

a numeric vector

trans

a numeric vector

outflo

a numeric vector

gini

a numeric vector

countryid

a character vector

References

Beckfield, Jason. 2006. "European integration and income inequality."" American Sociological Review 71(6): 964-985. Schoon, Eric W., David Melamed, and Ronald L. Breiger. 2024. Regression Inside Out. NY: Cambridge University Press.

Examples

data(Beckfield06)
head(Beckfield06)

Subset of data from the General Social Survey from 2016. Data were analyzed in Schoon, Melamed, and Breiger (2024).

Description

Subset of data from the General Social Survey from 2016. Data were analyzed in Schoon, Melamed, and Breiger (2024). Full details on the variable selection and source information is available therein.

Usage

data("GSS.2016")

Format

A data frame with 2867 observations on the following 27 variables.

sclass

a numeric vector

fulltime

a numeric vector

retired

a numeric vector

hrsworked

a numeric vector

occprestige

a numeric vector

occprestige_partner

a numeric vector

occprestige_mother

a numeric vector

occprestige_father

a numeric vector

children

a numeric vector

age

a numeric vector

educ

a numeric vector

paeduc

a numeric vector

maeduc

a numeric vector

speduc

a numeric vector

babs

a numeric vector

female

a numeric vector

white

a numeric vector

black

a numeric vector

other

a numeric vector

income

a numeric vector

republican

a numeric vector

conservative

a numeric vector

environment

a numeric vector

helpblackpeople

a numeric vector

science

a numeric vector

govequalwealth

a numeric vector

pclass

a numeric vector

References

Schoon, Eric W., David Melamed, and Ronald L. Breiger. 2024. Regression Inside Out. NY: Cambridge University Press.

Examples

data(GSS.2016)
head(GSS.2016)

Subset of the General Social Survey analyzed by Schoon, Melamed, and Breiger (2024)

Description

Subset of the General Social Survey analyzed by Schoon, Melamed, and Breiger (2024). Full details on the variable selection and source information is available therein.

Usage

data("GSS2018")

Format

A data frame with 558 observations on the following 7 variables.

dog

a numeric vector

race

a numeric vector

sex

a numeric vector

children

a numeric vector

married

a numeric vector

age

a numeric vector

income

a numeric vector

References

Schoon, Eric W., David Melamed, and Ronald L. Breiger. 2024. Regression Inside Out. NY: Cambridge University Press.

Examples

data(GSS2018)
head(GSS2018)

Replication data for regression models with a count dependent variable.

Description

Data analyzed by Hilbe (2011), and used here to illustrate model visualization and coefficient decomposition for count models.

Usage

data("Hilbe")

Format

A data frame with 601 observations on the following 9 variables.

naffairs

a numeric vector

avgmarr

a numeric vector

hapavg

a numeric vector

vryhap

a numeric vector

smerel

a numeric vector

vryrel

a numeric vector

yrsmarr4

a numeric vector

yrsmarr5

a numeric vector

yrsmarr6

a numeric vector

Source

Hilbe, Joseph M., 2011. Negative binomial regression. NY: Cambridge University Press.

Examples

data(Hilbe)
head(Hilbe)

Data to replicate OLS regression models reported in Kenworthy (1999).

Description

Data to replicate OLS regression models reported in Kenworthy (1999). Data were analyzed in Schoon, Melamed, and Breiger (2024). Full details on the variable selection and source information is available therein.

Usage

data("Kenworthy99")

Format

A data frame with 15 observations on the following 6 variables.

dv

a numeric vector

gdp

a numeric vector

pov

a numeric vector

tran

a numeric vector

ISO3

a character vector

nation.long

a character vector

References

Kenworthy, Lane. 1999. "Do social-welfare policies reduce poverty? A cross-national assessment."" Social Forces 77(3): 1119-1139. Schoon, Eric W., David Melamed, and Ronald L. Breiger. 2024. Regression Inside Out. NY: Cambridge University Press.

Examples

data(Kenworthy99)
head(Kenworthy99)

Subset of replication data from Ragin and Fiss (2017).

Description

Subset of replication data from Ragin and Fiss (2017). Data were analyzed in Schoon, Melamed, and Breiger (2024). Full details on the variable selection and source information is available therein.

Usage

data("RaginData")

Format

A data frame with 4185 observations on the following 10 variables.

incrat

a numeric

pinc

a numeric

ped

a numeric

resp_ed

a numeric

afqt

a numeric

kids

a numeric

married

a numeric

black

a numeric

male

a numeric

povd

a numeric

References

Ragin, Charles C. and Peer C. Fiss. 2017. Intersectional inequality: Race, class, test scores, and poverty. Chicago, IL: University of Chicago Press. Schoon, Eric W., David Melamed, and Ronald L. Breiger. 2024. Regression Inside Out. NY: Cambridge University Press.

Examples

data(RaginData)
head(RaginData)

Subset of replication data from Schneider and Makszin (2014).

Description

Subset of replication data from Schneider and Makszin (2014). Data were analyzed in Schoon, Melamed, and Breiger (2024). Full details on the variable selection and source information is available therein.

Usage

data("SchneiderAndMakszin06")

Format

A data frame with 30 observations on the following 36 variables.

id

a character vector

country

a character vector

year

a numeric vector

fde

a numeric vector

fde_cilb

a numeric vector

fde_ciub

a numeric vector

wcoord

a numeric vector

govint

a numeric vector

ud

a numeric vector

epl

a numeric vector

socexp

a numeric vector

eduexp

a numeric vector

vet_un

a numeric vector

lmexp

a numeric vector

wagecov

a numeric vector

vet_isced3

a numeric vector

eduexp_pri

a numeric vector

edu_terenr

a numeric vector

vt_reg

a numeric vector

vt_vap

a numeric vector

compvote

a numeric vector

fde2

a numeric vector

low_fde_l

a numeric vector

high_fde_l

a numeric vector

high_wc_l

a numeric vector

high_int_l

a numeric vector

high_ud_l

a numeric vector

high_epl_l

a numeric vector

high_socx_l

a numeric vector

high_edux_l

a numeric vector

high_lmx_l

a numeric vector

high_vet_l

a numeric vector

p1_y

a numeric vector

p2_y

a numeric vector

p3_y

a numeric vector

sol_y

a numeric vector

References

Schneider, Carsten Q., and Kristin Makszin. 2014. "Forms of welfare capitalism and education-based participatory inequality." Socio-Economic Review 12(2): 437-462. Schoon, Eric W., David Melamed, and Ronald L. Breiger. 2024. Regression Inside Out. NY: Cambridge University Press.

Examples

data(SchneiderAndMakszin06)
head(SchneiderAndMakszin06)

Subset of replication data from Wimmer, Cederman, and Min (2009).

Description

Subset of replication data from Wimmer, Cederman, and Min (2009). Data were analyzed in Schoon, Melamed, and Breiger (2024). Full details on the variable selection and source information is available therein.

Usage

data("Wimmer_et_al_EPR")

Format

A data frame with 7908 observations on the following 80 variables.

yearc

a numeric

year

a numeric

cowcode

a numeric

country

a character

gdpcap

a numeric

gdpcapl

a numeric

oilpc

a numeric

oilpcl

a numeric

popavg

a numeric

lpopl

a numeric

ethfrac

a numeric

western

a numeric

eeurop

a numeric

lamerica

a numeric

ssafrica

a numeric

asia

a numeric

nafrme

a numeric

lmtnest

a numeric

polity2

a numeric

polity

a numeric

anoc

a numeric

anocl

a numeric

democ

a numeric

democl

a numeric

regchg3

a numeric

pimppast

a numeric

groups

a numeric

egipgrps

a numeric

exclgrps

a numeric

exclpop

a numeric

lrexclpop

a numeric

ttlpop

a numeric

discpop

a numeric

pwrlpop

a numeric

olppop

a numeric

olpspop

a numeric

jppop

a numeric

sppop

a numeric

dompop

a numeric

monpop

a numeric

maxexclpop

a numeric

maxegippop

a numeric

maxpop

a numeric

newonset

a numeric

newethonset

a numeric

newhionset

a numeric

newethhionset

a numeric

onsetstatus

a numeric

onsetstatus2

a numeric

actoraim

a numeric

actoraim2

a numeric

ongoingwarl

a numeric

ongoinghiwarl

a numeric

newonset2

a numeric

newhionset2

a numeric

newethonset2

a numeric

warlfl

a numeric

onsetfl

a numeric

ethonsetfl

a numeric

onsetfl2

a numeric

ethonsetfl2

a numeric

warstns2

a numeric

warstns1

a numeric

atwarnsl

a numeric

npeaceyears

a numeric

nspline1

a numeric

nspline2

a numeric

nspline3

a numeric

hpeaceyears

a numeric

hspline1

a numeric

hspline2

a numeric

hspline3

a numeric

fpeaceyears

a numeric

fspline1

a numeric

fspline2

a numeric

fspline3

a numeric

speaceyears

a numeric

sspline1

a numeric

sspline2

a numeric

sspline3

a numeric

References

Wimmer, Andreas, Lars-Erik Cederman, and Brian Min. 2009. "Ethnic politics and armed conflict: A configurational analysis of a new global data set." American Sociological Review 74(2): 316-337.

Examples

data(Wimmer_et_al_EPR)
head(Wimmer_et_al_EPR)

Compute the Cosine similarity between two points.

Description

Given two points, the function computes the cosine similarity between them.

Usage

cosine(x,y)

Arguments

x

Point 1

y

Point 2

Value

The cosine similarity, ranging between -1 and +1.

Author(s)

Ronald L. Breiger, David Melamed and Eric Schoon

References

Schoon, Eric, David Melamed, and Ronald L. Breiger. 2023. Regression Inside Out. NY: Cambridge University Press.

Examples

data(Kenworthy99)
m1 <- lm(scale(dv) ~ scale(gdp) + scale(pov) + scale(tran) -1,data=Kenworthy99)
rp1 <- rio.plot(m1,include.int="no",r1=1:15)
cosine(rp1$row.dimensions[15,],rp1$row.dimensions[8,]) 
# cosine similarity between USA and Ireland

cosine(rp1$row.dimensions[15,],rp1$row.dimensions[14,]) 
# cosine similarity between USA and United Kingdom

Decompose the Results of a Regression Model by Cases

Description

This function takes a regression model object and a vector of case assignments to groups (note, cases can be in their own group) and computes each cases' contribution to the overall regression coefficients.

Usage

decompose.model(m1,group.by=group.by,include.int="yes",model.type="OLS")

Arguments

m1

A regression model object. OLS, logistic, Poisson and negative binomial regression are supported.

group.by

A numeric vector denoting group membership. Should be the same length as the number of cases.

include.int

Whether the regression model included an intercept. Default is "yes."

model.type

Type of model to be decomposed. OLS via lm, logistic via glm ("logit"), Poisson via glm ("poisson"), and negative binomial via MASS ("nb") are supported.

Value

decomp.coef

Each case's or subset of cases' contribution to the estimated slope or regression coefficient.

decomp.var

Each case's or subset of cases' contribution to the variance of the estimated slope or regression coefficient.

Author(s)

David Melamed, Ronald L. Breiger, and Eric Schoon

References

Schoon, Eric, David Melamed, and Ronald L. Breiger. 2024. Regression Inside Out. NY: Cambridge University Press.

Examples

data(Kenworthy99)
m1 <- lm(scale(dv) ~ scale(gdp) + scale(pov) + scale(tran) -1,data=Kenworthy99)
decompose.model(m1,group.by=c("Liberal","Corp","Liberal",
"SocDem","SocDem","Corp","Corp","Corp","Corp","Corp","SocDem",
"SocDem","Liberal","Liberal","Liberal"),include.int="no")

Project point 1 onto the line (at 90 degress) running through point 2 and the origin (0,0).

Description

Given two points, p1 and p2, this function identifies the point at which p1 is projected onto the line connecting p2 and the origin (0,0). The projection occurs at a right angle.

Usage

project.point(p1,p2)

Arguments

p1

First point, the one that is to be projected onto point 2.

p2

Second point, the one that is projected to the origin. This is the outcome or dependent variable in our book. See reference below.

Details

The output is just a single point. This is implemented as the point to which lines are drawn in many graphs.

Value

Two values which correspond to the x and y co-ordinates in the graph.

Author(s)

David Melamed, Ronald L. Breiger, and Eric Schoon

References

Schoon, Eric, David Melamed, and Ronald L. Breiger. 2024. Regression Inside Out. NY: Cambridge University Press.

Examples

data(Kenworthy99)
m1 <- lm(scale(dv) ~ scale(gdp) + scale(pov) + scale(tran) -1,data=Kenworthy99)
rp1 <- rio.plot(m1,include.int="no",r1=1:15)
project.point(as.numeric(rp1$col.dimensions[1,]),as.numeric(rp1$row.dimensions[1,]))

Regression Inside Out: Plotting Regression Models

Description

rio.plot is used to generate a reduced rank image of a regression model. The function computes row and column dimensions for both cases and variables, and generates an image of the model based on those scores.

Usage

rio.plot(m1,exclude.vars="no",r1="none",case.names="",col.names="no",
h.just=-.2,v.just=0,case.col="blue",var.name.col="black",
include.int="yes",group.cases=1,model.type="OLS")

Arguments

m1

a regression model object. Supported models include OLS, Logistic, Poisson, and Negative Binomial Regression.

exclude.vars

an optional numerical vector indicating variables from the model to exclude from the plot of the model.

r1

an optional numerical vector indicating cases to include in the plot. By default, all cases are excluded from the plot.

case.names

a character string of names to label the cases. Should be the same length as 'r1.'

col.names

whether to include the variable names in the plot. Default is "no"

h.just

horizontal justification in the plot. Default is -.2

v.just

vertical justification in the plot. Default is 0

case.col

if cases are added to the plot, this is their color. Default is "blue"

var.name.col

Color of the names of variables in the plot. Default is "black"

include.int

Whether the underlying model included a model intercept. Default is "yes"

group.cases

Whether to aggregate cases into clusters or subsets. If yes, provide a numeric vector of memberships. It will aggregate over them by summing.

model.type

The type of regression model. OLS is supported via the lm function. Logistic and Poisson regression are supported via the glm function. Negative Binomial regression is supported via the MASS package. Default is "OLS." For logistic regression, use "logit." For Poisson regression, use "poisson." For negative binomial regression, use "nb."

Details

The function take a regression model object (OLS, logistic, Poisson, or negative binomial) and computes the corresponding row (case) and column (variables) scores. The scores are part of the output, as is a ggplot object of the model.

Value

rio.plot returns several objects.

p1

a ggplot object of the model space, given the terms in the function

row.dimensions

the scores assigned to each case, or each subset of cases if they were aggregated using the 'group.cases' option. These are the co-ordinates in the plot.

col.dimensions

the scores assigned to each variable. These are the co-ordinates in the plot.

case.variances

each cases' contribution (or each subsets' contribution) to the variance of the estimated regression coefficient

U

The orthogonalized column space matrix from the Singular Value Decomposition of the predictor matrix and fitted values.

UUt

The orthogonalized column space matrix from the Singular Value Decomposition of the predictor matrix and fitted values, post-multiplied by its transpose.

Author(s)

David Melamed, Ronald L. Breiger, and Eric Schoon

References

Schoon, Eric, David Melamed, and Ronald L. Breiger. 2024. Regression Inside Out. NY: Cambridge University Press.

Examples

data(Kenworthy99)
m1 <- lm(scale(dv) ~ scale(gdp) + scale(pov) + scale(tran) -1,data=Kenworthy99)
rp1 <- rio.plot(m1,include.int="no")
names(rp1)
rp1$gg.obj 
# rp1$gg.obj + ggplot2::scale_x_continuous(limits=c(-.55,1)) # useful option

rp2 <- rio.plot(m1,r1=1:15,case.names=paste(1:15),include.int="no")
rp2$gg.obj

Kenworthy99 <- data.frame(Kenworthy99,type=c("Liberal","Corp","Liberal",
"SocDem","SocDem","Corp","Corp","Corp","Corp","Corp","SocDem","SocDem",
"Liberal","Liberal","Liberal"))

rp3 <- rio.plot(m1,r1=1:15,group.cases=Kenworthy99$type,include.int="no")
rp3$gg.obj 
# rp3$gg.obj + ggplot2::scale_x_continuous(limits=c(-1,20))