Show Answer
#remotes::install_github("leeper/cregg")
library(gtools)
library(tidyverse)
library(xtable)
library(cjoint)
library(cregg)
library(sjPlot)
library(ggplot2)
library(graphics)
library(scales)
library(estimatr)
set.seed(12345678)Conjoint Experiment
1 Replication Task
In this exercise, we will replicate the study by O’Campo-Roland. The article’s research question guiding the conjoint design is whether and how white and Black Americans differ in their perceptions of which Latinos are considered “American.” The design systematically varies Latino characteristics such as phenotype, immigrant background/legal status, language, socioeconomic markers, religion, and political affiliation. The goal is to uncover how prototypical (white) versus peripheral (Black) group members draw national boundaries of inclusion, and which Latino subgroups are deemed part of the American imagined community.
The study employs a paired-profile conjoint survey experiment to assess perceptions of Latinos’ Americanness. A sample of 750 white and 750 Black respondents (total N=1,500) was recruited through YouGov. Each respondent evaluated ten Latino profiles, presented in five side-by-side tasks, and for each task was asked (1) to choose which profile they thought was “more American” and (2) to rate each profile on a 1–7 Americanness scale.
The profiles were randomly generated by varying a set of attributes that capture Latino heterogeneity, including:
Physical appearance (photos from the Chicago Face Database spanning skin tones and genders);
Immigrant background/legal status (undocumented, legal, naturalized, or U.S.-born descendants of legal/undocumented immigrants);
Language ability (only English, bilingual, limited English, no English);
Socioeconomic characteristics (education and occupation levels);
Ethnic/national identification (American, Latino, Mexican, etc.);
Religion (Catholic, Evangelical, Presbyterian, not religious);
Political affiliation (Democrat, Republican, Independent, non-political);
Spousal background (American, white, Black, same as profile)
Design restrictions prevented unrealistic combinations (e.g., inconsistent education and occupation pairings). Analytically, the study estimated Average Marginal Component Effects (AMCEs) across all respondents, then compared marginal mean differences and attribute salience between white and Black respondents, as well as variation within groups by strength of American or racial identity. We will replicate the article findings on the main analysis.
1.1 Instaling necessary packages and functions
1.2 Data Cleaning
Download the data set.
Show Answer
###SET WORKING DIRECTORY and LOAD FILES ----
onedrive_url <- "https://1drv.ms/x/c/0f7d0a7f95ac8880/Edv5Acp5lARJicU-cxAi6q0B9Zn36gXoE4Syl_gv-8MCLw?e=x3jCEe&download=1"
#full <- read.csv(url(onedrive_url))
full <- read.csv('C:/Users/jmcar/OneDrive/Documentos/JOAO/UNC/2025 Fall/Advanced Causal/Class 1/Replication Material OCampo 2025/b_data/1_raw_data_for_long.csv')
##### Recoding key analytic variables----
full <- full %>%
mutate(#Race of respondent
race =
case_when(race == 1 ~ 'White respondent',
race == 2 ~ 'Black respondent'),
race = factor(race, levels = c("White respondent", "Black respondent")),
#American identity importance
amerimport =
case_when(Q7_2 == 1 ~ 1,
Q7_2 == 2 ~ 2,
Q7_2 == 3 ~ 3,
Q7_2 == 4 ~ 4),
#Racial identity importance
raceimport =
case_when(Q7_3 == 1 ~ 1,
Q7_3 == 2 ~ 2,
Q7_3 == 3 ~ 3,
Q7_3 == 4 ~ 4),
#Perceived prototypicality
prototypical =
case_when(Q7_4 == 1 ~ 1,
Q7_4 == 2 ~ 2,
Q7_4 == 3 ~ 3,
Q7_4 == 4 ~ 4))
###Splitting identity perception variables into two and recoding the variables into high and low ----
full$amerimport_high <- quantcut(full$amerimport, q=2)
full$raceimport_high <- quantcut(full$raceimport, q=2)
full$prototypical_high <- quantcut(full$prototypical, q=2)
full <- full %>%
mutate(amerimport_high =
case_when(amerimport_high == "[1,3]" ~ 'Low',
amerimport_high == "(3,4]" ~ 'High'),
raceimport_high =
case_when(raceimport_high == "[1,3]" ~ 'Low',
raceimport_high == "(3,4]" ~ 'High'),
prototypical_high =
case_when(prototypical_high == "[1,3]" ~ 'Low',
prototypical_high == "(3,4]" ~ 'High'))1.3 Reshaping Data
In order to perform the conjoint analysis, the data frame must be in a long format, which means that each row will contain a task, instead of a respondent itself. The function cj_tidy from the package cregg can help with it. We just need to specify a list containing the characteristics of each task and a list with the outcome.
Show Answer
####Reshaping data into long form ----
full <- full %>%
rename (choice_a = Q3_2,
choice_b = Q3_5,
choice_c = Q3_8,
choice_d = Q3_11,
choice_e = Q3_14,
rating_a1 = Q3_3_1,
rating_a2 = Q3_3_2,
rating_b1 = Q3_6_1,
rating_b2 = Q3_6_2,
rating_c1 = Q3_9_1,
rating_c2 = Q3_9_2,
rating_d1 = Q3_12_1,
rating_d2 = Q3_12_2,
rating_e1 = Q3_15_1,
rating_e2 = Q3_15_2)
# Characteriscs list
list1 <- list(
Faces = list(
~ Q3A_1_Faces + Q3B_1_Faces + Q3C_1_Faces + Q3D_1_Faces + Q3E_1_Faces,
~ Q3A_2_Faces + Q3B_2_Faces + Q3C_2_Faces + Q3D_2_Faces + Q3E_2_Faces
),
Background = list(
~ Q3A_1_Background + Q3B_1_Background + Q3C_1_Background + Q3D_1_Background + Q3E_1_Background,
~ Q3A_2_Background + Q3B_2_Background + Q3C_2_Background + Q3D_2_Background + Q3E_2_Background
),
Identifies = list(
~ Q3A_1_Identifies + Q3B_1_Identifies + Q3C_1_Identifies + Q3D_1_Identifies + Q3E_1_Identifies,
~ Q3A_2_Identifies + Q3B_2_Identifies + Q3C_2_Identifies + Q3D_2_Identifies + Q3E_2_Identifies
),
Language = list(
~ Q3A_1_Language + Q3B_1_Language + Q3C_1_Language + Q3D_1_Language + Q3E_1_Language,
~ Q3A_2_Language + Q3B_2_Language + Q3C_2_Language + Q3D_2_Language + Q3E_2_Language
),
Education = list(
~ Q3A_1_Education + Q3B_1_Education + Q3C_1_Education + Q3D_1_Education + Q3E_1_Education,
~ Q3A_2_Education + Q3B_2_Education + Q3C_2_Education + Q3D_2_Education + Q3E_2_Education
),
Profession = list(
~ Q3A_1_Profession + Q3B_1_Profession + Q3C_1_Profession + Q3D_1_Profession + Q3E_1_Profession,
~ Q3A_2_Profession + Q3B_2_Profession + Q3C_2_Profession + Q3D_2_Profession + Q3E_2_Profession
),
Religion = list(
~ Q3A_1_Religion + Q3B_1_Religion + Q3C_1_Religion + Q3D_1_Religion + Q3E_1_Religion,
~ Q3A_2_Religion + Q3B_2_Religion + Q3C_2_Religion + Q3D_2_Religion + Q3E_2_Religion
),
Political = list(
~ Q3A_1_Political + Q3B_1_Political + Q3C_1_Political + Q3D_1_Political + Q3E_1_Political,
~ Q3A_2_Political + Q3B_2_Political + Q3C_2_Political + Q3D_2_Political + Q3E_2_Political
),
Spouse = list(
~ Q3A_1_Spouse + Q3B_1_Spouse + Q3C_1_Spouse + Q3D_1_Spouse + Q3E_1_Spouse,
~ Q3A_2_Spouse + Q3B_2_Spouse + Q3C_2_Spouse + Q3D_2_Spouse + Q3E_2_Spouse
),
rating = list(
~ rating_a1 + rating_b1 + rating_c1 + rating_d1 + rating_e1,
~ rating_a2 + rating_b2 + rating_c2 + rating_d2 + rating_e2
)
)
# Response list
list2 <- list(choice = ~ choice_a + choice_b + choice_c + choice_d + choice_e)
# Aplying the function
long <- cregg::cj_tidy(full,
profile_variables = list1,
task_variables = list2,
id = ~ caseid)
# Need to save and reload the data in order for code to run
write.csv(long, 'long.csv')
long <- read.csv('long.csv')After reshaping the data set, we will recode the categories of each group in meaningfull terms, transforming the numerical codes into actual categories.
Show Answer
# Recoding profile characteristics as factors
long <- long %>%
mutate(
# Ensure numeric codes (or character across-the-board)
across(c(Faces, Background, Identifies, Language, Education, Profession, Religion, Political, Spouse),
~ as.integer(.x)),
Faces = factor(recode(Faces,
`1`="White man", `2`="White woman", `3`="Light man", `4`="Light woman",
`5`="Medium man", `6`="Medium woman", `7`="Brown man", `8`="Brown woman",
`9`="Black man", `10`="Black woman",
.default = NA_character_
), levels = c("White man","White woman","Light man","Light woman",
"Medium man","Medium woman","Brown man","Brown woman",
"Black man","Black woman"), ordered = TRUE),
Background = factor(recode(Background,
`1`="Legal immigrant", `2`="Undoc. immigrant", `3`="Naturalized citizen",
`4`="US-born legal parents", `5`="US-born undoc. parents",
`6`="US-born legal grandparents", `7`="US-born undoc. grandparents",
.default = NA_character_
), levels = c("Legal immigrant","Undoc. immigrant","Naturalized citizen",
"US-born legal parents","US-born undoc. parents",
"US-born legal grandparents","US-born undoc. grandparents")),
Identifies = factor(recode(Identifies,
`1`="American", `2`="Latino", `3`="Hispanic", `4`="Mexican-American",
`5`="Mexican", `6`="White", `7`="Black", .default = NA_character_
), levels = c("American","Latino","Hispanic","Mexican-American","Mexican","White","Black")),
Language = factor(recode(Language,
`1`="Only English", `2`="Bilingual", `3`="Limited English", `4`="No English",
.default = NA_character_
), levels = c("Only English","Bilingual","Limited English","No English"),
ordered = TRUE),
Education = factor(recode(Education,
`1`="High school dropout", `2`="High school graduate",
`3`="Bachelor's degree", `4`="Advanced/graduate degree",
.default = NA_character_
), levels = c("High school dropout","High school graduate",
"Bachelor's degree","Advanced/graduate degree"),
ordered = TRUE),
Profession = factor(recode(Profession,
`1`="Waiter", `2`="Janitor", `3`="Sales manager", `4`="Teacher",
`5`="IT professional", `6`="Doctor", `7`="Small business owner",
.default = NA_character_
), levels = c("Waiter","Janitor","Sales manager","Teacher",
"IT professional","Doctor","Small business owner")),
Religion = factor(recode(Religion,
`1`="Presbyterian", `2`="Evangelical", `3`="Catholic", `4`="Not religious",
.default = NA_character_
), levels = c("Presbyterian","Evangelical","Catholic","Not religious")),
Political = factor(recode(Political,
`1`="Democrat", `2`="Republican", `3`="Independent", `4`="Not political",
.default = NA_character_
), levels = c("Democrat","Republican","Independent","Not political")),
Spouse = factor(recode(Spouse,
`1`="Not married", `2`="S. is American", `3`="S. is Black",
`4`="S. is White", `5`="S. is same as profile",
.default = NA_character_
), levels = c("Not married","S. is American","S. is Black",
"S. is White","S. is same as profile"))
)
# cleaning American rating variable
df <- long %>%
mutate(rating =
case_when(rating == 1 ~ 1,
rating == 2 ~ 2,
rating == 3 ~ 3,
rating == 4 ~ 4,
rating == 5 ~ 5,
rating == 6 ~ 6,
rating == 7 ~ 7))
df$rating_r <- rescale(df$rating) #rescales American ratings from 0 to 1
tab_df(df |> slice_head(n= 10))| X | caseid | weight | race | educ | gender | employ | faminc_new | religpew | birthyr | ideo5 | pid7 | presvote16post | votereg | Q7_2 | Q7_3 | Q7_4 | Q4_3 | Q4_5_2 | amerimport | raceimport | prototypical | amerimport_high | raceimport_high | prototypical_high | task | profile | Faces | Background | Identifies | Language | Education | Profession | Religion | Political | Spouse | rating | choice | pair | rating_r |
| 1.10 | 1203595341 | 1.55 | White respondent | 5 | 1 | 1 | 6 | 1 | 1963 | 4 | 6 | 2 | 1 | 3 | 2 | 3 | 1 | 1 | 3 | 2 | 3 | Low | Low | Low | 1 | A | Brown woman | Undoc. immigrant | Hispanic | Limited English | High school dropout | Waiter | Presbyterian | Not political | S. is same as profile | 2 | 2 | 1 | 0.17 |
| 2.10 | 1204107895 | 1.03 | White respondent | 2 | 2 | 7 | 1 | 1 | 1988 | 6 | 8 | 7 | 2 | 1 | 1 | 2 | 1 | 1 | 1 | 1 | 2 | Low | Low | Low | 1 | A | Medium man | Undoc. immigrant | Hispanic | Bilingual | Bachelor's degree | Janitor | Presbyterian | Independent | Not married | 5 | 1 | 2 | 0.67 |
| 3.10 | 1204107941 | 0.76 | White respondent | 1 | 1 | 6 | 6 | 1 | 1953 | 6 | 5 | 7 | 2 | 4 | 4 | 4 | 1 | 1 | 4 | 4 | 4 | High | High | High | 1 | A | Medium woman | US-born legal parents | American | Only English | Bachelor's degree | Small business owner | Presbyterian | Republican | S. is same as profile | 5 | 1 | 3 | 0.67 |
| 4.10 | 1204110451 | 0.90 | White respondent | 3 | 2 | 1 | 4 | 9 | 1977 | 2 | 3 | 1 | 1 | 2 | 2 | 3 | 1 | 1 | 2 | 2 | 3 | Low | Low | Low | 1 | A | Brown man | US-born legal parents | Mexican | Only English | High school dropout | Sales manager | Not religious | Independent | S. is Black | 7 | 1 | 4 | 1.00 |
| 5.10 | 1204116055 | 0.72 | White respondent | 6 | 1 | 8 | 6 | 1 | 1994 | 3 | 4 | 5 | 1 | 2 | 2 | 3 | 1 | 1 | 2 | 2 | 3 | Low | Low | Low | 1 | A | Brown woman | US-born undoc. grandparents | Mexican-American | Only English | High school graduate | Janitor | Presbyterian | Independent | S. is same as profile | 7 | 1 | 5 | 1.00 |
| 6.10 | 1204116247 | 1.17 | White respondent | 2 | 1 | 1 | 13 | 11 | 1966 | 4 | 7 | 2 | 1 | 2 | 1 | 4 | 1 | 1 | 2 | 1 | 4 | Low | Low | High | 1 | A | Black man | Undoc. immigrant | American | Bilingual | Bachelor's degree | Sales manager | Evangelical | Republican | Not married | 2 | 2 | 6 | 0.17 |
| 7.10 | 1204116147 | 0.79 | White respondent | 2 | 2 | 5 | 2 | 11 | 1955 | 2 | 4 | 7 | 2 | 4 | 4 | 4 | 1 | 1 | 4 | 4 | 4 | High | High | High | 1 | A | Medium man | US-born legal parents | Mexican-American | Only English | Advanced/graduate degree | Janitor | Not religious | Not political | S. is American | 2 | 1 | 7 | 0.17 |
| 8.10 | 1204116213 | 1.41 | White respondent | 4 | 1 | 1 | 4 | 2 | 1969 | 5 | 7 | 2 | 1 | 4 | 2 | 1 | 1 | 1 | 4 | 2 | 1 | High | Low | Low | 1 | A | White woman | US-born legal parents | American | Limited English | Bachelor's degree | Teacher | Evangelical | Not political | S. is Black | 6 | 1 | 8 | 0.83 |
| 9.10 | 1204117027 | 1.03 | White respondent | 4 | 2 | 6 | 1 | 9 | 1958 | 1 | 1 | 7 | 1 | 4 | 2 | 3 | 1 | 1 | 4 | 2 | 3 | High | Low | Low | 1 | A | Light woman | US-born undoc. grandparents | White | Limited English | Advanced/graduate degree | Teacher | Presbyterian | Independent | S. is same as profile | 7 | 2 | 9 | 1.00 |
| 10.10 | 1204116383 | 0.98 | White respondent | 3 | 1 | 5 | 2 | 2 | 1948 | 3 | 4 | 7 | 1 | 4 | 4 | 4 | 1 | 1 | 4 | 4 | 4 | High | High | High | 1 | A | White woman | Naturalized citizen | American | No English | High school dropout | Janitor | Evangelical | Independent | Not married | 6 | 1 | 10 | 0.83 |
To end, we will transform the data again so the factor categories are well defined within the conjoint desing. It is important to notice that, due to a package bug, we need to repeat this process exporting the previous file and downloading it back again.
Show Answer
write.csv(df, 'df.csv')
df <- read.csv('df.csv')
####Reshaping data as factors ----
df$race <- as.factor(df$race)
df <- df %>%
mutate(Faces = factor(Faces, levels = c("White man", "White woman", "Light man", "Light woman",
"Medium man", "Medium woman", "Brown man",
"Brown woman", "Black man", "Black woman"),
ordered = is.ordered(Faces)))
df <- df %>%
mutate(Background = factor(Background, levels = c("Legal immigrant",
"Undoc. immigrant",
"Naturalized citizen",
"US-born legal parents",
"US-born undoc. parents",
"US-born legal grandparents",
"US-born undoc. grandparents")))
df <- df %>%
mutate(Identifies = factor(Identifies, levels = c("American",
"Latino",
"Hispanic",
"Mexican-American",
"Mexican",
"White",
"Black")))
df <- df %>%
mutate(Language = factor(Language, levels = c("Only English",
"Bilingual",
"Limited English",
"No English"),
ordered = is.ordered(Language)))
df <- df %>%
mutate(Education = factor(Education, levels = c("High school dropout",
"High school graduate",
"Bachelor's degree",
"Advanced/graduate degree"),
ordered = is.ordered(Education)))
df <- df %>%
mutate(Profession = factor(Profession, levels = c("Waiter", "Janitor",
"Sales manager", "Teacher",
"IT professional", "Doctor",
"Small business owner")))
df <- df %>%
mutate(Religion = factor(Religion, levels = c("Presbyterian", "Evangelical",
"Catholic", "Not religious")))
df <- df %>%
mutate(Political = factor(Political, levels = c("Democrat", "Republican",
"Independent", "Not political")))
df <- df %>%
mutate(Spouse = factor(Spouse, levels = c("Not married",
"S. is American",
"S. is Black",
"S. is White",
"S. is same as profile")))1.4 Identify the conjoint design by OLS
Accounting for a simple conjoint design, the model can be identified with a simple OLS regression with clustered standard errors by respondent. We can easily estimate this using the lm_robust method from the estimatr package. We just need to specify the equation, the clusters and the weights within it.
Show Answer
fit_ols <- lm_robust(rating_r ~ Faces + Background + Identifies +
Language + Education + Profession + Religion
+ Political + Spouse,
data=df, clusters = caseid, weights = weight)
tab_model(fit_ols)| rating r | |||
| Predictors | Estimates | CI | p |
| (Intercept) | 0.77 | 0.73 – 0.80 | <0.001 |
| Faces [White woman] | -0.01 | -0.03 – 0.01 | 0.363 |
| Faces [Light man] | -0.01 | -0.03 – 0.02 | 0.607 |
| Faces [Light woman] | -0.01 | -0.03 – 0.01 | 0.379 |
| Faces [Medium man] | -0.01 | -0.03 – 0.01 | 0.398 |
| Faces [Medium woman] | -0.01 | -0.03 – 0.01 | 0.316 |
| Faces [Brown man] | -0.03 | -0.05 – -0.01 | 0.003 |
| Faces [Brown woman] | -0.02 | -0.04 – 0.00 | 0.082 |
| Faces [Black man] | -0.01 | -0.03 – 0.01 | 0.394 |
| Faces [Black woman] | 0.01 | -0.01 – 0.03 | 0.362 |
| Background [Undoc. immigrant] |
-0.16 | -0.19 – -0.14 | <0.001 |
| Background [Naturalized citizen] |
0.02 | -0.00 – 0.04 | 0.053 |
| Background [US-born legal parents] |
0.05 | 0.03 – 0.07 | <0.001 |
| Background [US-born undoc. parents] |
-0.01 | -0.03 – 0.01 | 0.330 |
| Background [US-born legal grandparents] |
0.03 | 0.02 – 0.05 | <0.001 |
| Background [US-born undoc. grandparents] |
0.01 | -0.01 – 0.03 | 0.489 |
| Identifies [Latino] | -0.02 | -0.04 – -0.00 | 0.032 |
| Identifies [Hispanic] | -0.02 | -0.04 – -0.00 | 0.018 |
| Identifies [Mexican-American] |
-0.02 | -0.03 – 0.00 | 0.055 |
| Identifies [Mexican] | -0.03 | -0.04 – -0.01 | 0.005 |
| Identifies [White] | -0.01 | -0.03 – 0.02 | 0.616 |
| Identifies [Black] | -0.01 | -0.03 – 0.02 | 0.600 |
| Language [Bilingual] | -0.01 | -0.02 – 0.01 | 0.447 |
| Language [Limited English] |
-0.04 | -0.05 – -0.02 | <0.001 |
| Language [No English] | -0.06 | -0.09 – -0.04 | <0.001 |
| Education [High school graduate] |
0.00 | -0.01 – 0.02 | 0.730 |
| Education [Bachelor's degree] |
0.02 | 0.00 – 0.03 | 0.033 |
| Education [Advanced/graduate degree] |
0.01 | -0.00 – 0.03 | 0.113 |
| Profession [Janitor] | -0.01 | -0.03 – 0.00 | 0.160 |
| Profession [Sales manager] |
-0.00 | -0.02 – 0.01 | 0.631 |
| Profession [Teacher] | -0.01 | -0.03 – 0.01 | 0.361 |
| Profession [IT professional] |
0.00 | -0.02 – 0.02 | 0.711 |
| Profession [Doctor] | 0.00 | -0.02 – 0.03 | 0.900 |
| Profession [Small business owner] |
0.00 | -0.01 – 0.02 | 0.522 |
| Religion [Evangelical] | -0.00 | -0.02 – 0.01 | 0.708 |
| Religion [Catholic] | 0.00 | -0.01 – 0.02 | 0.545 |
| Religion [Not religious] | -0.00 | -0.01 – 0.01 | 0.828 |
| Political [Republican] | -0.00 | -0.02 – 0.01 | 0.575 |
| Political [Independent] | 0.01 | -0.01 – 0.02 | 0.472 |
| Political [Not political] | -0.01 | -0.02 – 0.01 | 0.248 |
| Spouse [S. is American] | 0.00 | -0.01 – 0.02 | 0.617 |
| Spouse [S. is Black] | -0.01 | -0.02 – 0.01 | 0.335 |
| Spouse [S. is White] | -0.01 | -0.02 – 0.01 | 0.458 |
| Spouse [S. is same as profile] |
-0.00 | -0.02 – 0.01 | 0.467 |
| Observations | 14979 | ||
| R2 / R2 adjusted | 0.060 / 0.057 | ||
1.5 Defining the conjoint design
However, we can incorporate more characteristics of the design by using specific packages. For instance, we can run the function amce from the package cjoint to specify design restrictions, such as impossible combinations and a relative probability of each category to be presented.
Show Answer
###CONJOINT DESIGN ----
# Providing a list with all the attributes available by each feature
attribute_list <- list()
attribute_list[["Faces"]] <-c("White man", "White woman", "Light man", "Light woman",
"Medium man", "Medium woman", "Brown man", "Brown woman",
"Black man", "Black woman")
attribute_list[["Background"]] <- c("Legal immigrant", "Undoc. immigrant",
"Naturalized citizen",
"US-born legal parents",
"US-born undoc. parents",
"US-born legal grandparents",
"US-born undoc. grandparents")
attribute_list[["Identifies"]] <- c("American", "Latino",
"Hispanic", "Mexican-American",
"Mexican", "White", "Black")
attribute_list[["Language"]] <- c("Only English",
"Bilingual",
"Limited English",
"No English")
attribute_list[["Education"]] <- c("High school dropout",
"High school graduate",
"Bachelor's degree",
"Advanced/graduate degree")
attribute_list[["Profession"]] <- c("Waiter", "Janitor",
"Sales manager", "Teacher",
"IT professional", "Doctor",
"Small business owner")
attribute_list[["Religion"]] <- c("Presbyterian", "Evangelical",
"Catholic", "Not religious")
attribute_list[["Political"]] <- c("Democrat", "Republican",
"Independent", "Not political")
attribute_list[["Spouse"]] <- c("Not married", "S. is American",
"S. is Black", "S. is White",
"S. is same as profile")
# Providing a list with all the impossible attributes to be combined
# (example a doctor that don't have highschool)
constraint_list <- list()
constraint_list[[1]] <- list()
constraint_list[[1]][["Education"]] <- c("High school dropout",
"High school graduate")
constraint_list[[1]][["Profession"]] <- c("Doctor",
"IT professional", "Teacher")
constraint_list[[2]] <- list()
constraint_list[[2]][["Education"]] <- c("Bachelor's degree")
constraint_list[[2]][["Profession"]] <- c("Doctor")
constraint_list[[3]] <- list()
constraint_list[[3]][["Background"]] <- c("US-born legal parents",
"US-born undoc. parents",
"US-born legal grandparents",
"US-born undoc. grandparents")
constraint_list[[3]][["Language"]] <- c("No English")
constraint_list[[4]] <- list()
constraint_list[[4]][["Background"]] <- c("Legal immigrant",
"Naturalized citizen",
"Undoc. immigrant")
constraint_list[[4]][["Language"]] <- c("Only English")
constraint_list[[5]] <- list()
constraint_list[[5]][["Profession"]] <- c("IT professional",
"Doctor",
"Sales manager",
"Teacher")
constraint_list[[5]][["Language"]] <- c("No English")
constraint_list[[6]] <- list()
constraint_list[[6]][["Faces"]] <- c("White man", "White woman",
"Light man", "Light woman",
"Medium man", "Medium woman")
constraint_list[[6]][["Identifies"]] <- c("Black")
constraint_list[[7]] <- list()
constraint_list[[7]][["Faces"]] <- c("Brown man", "Brown woman",
"Black man", "Black woman")
constraint_list[[7]][["Identifies"]] <- c("White")
# To end, we need to provide a list with the
marginal_weights <- list()
marginal_weights[["Faces"]] <- rep(1/length(attribute_list[["Faces"]]),
length(attribute_list[["Faces"]]))
marginal_weights[["Background"]] <- rep(1/length(attribute_list[["Background"]]),
length(attribute_list[["Background"]]))
marginal_weights[["Identifies"]] <- rep(1/length(attribute_list[["Identifies"]]),
length(attribute_list[["Identifies"]]))
marginal_weights[["Language"]] <- rep(1/length(attribute_list[["Language"]]),
length(attribute_list[["Language"]]))
marginal_weights[["Education"]] <- rep(1/length(attribute_list[["Education"]]),
length(attribute_list[["Education"]]))
marginal_weights[["Profession"]] <- rep(1/length(attribute_list[["Profession"]]),
length(attribute_list[["Profession"]]))
marginal_weights[["Religion"]] <- rep(1/length(attribute_list[["Religion"]]),
length(attribute_list[["Religion"]]))
marginal_weights[["Political"]] <- c(2/9, 2/9, 2/9, 3/9)
marginal_weights[["Spouse"]] <- c(5/15, 3/15, 2/15, 2/15, 3/15)
conjointdesign <- cjoint::makeDesign(type='constraints', attribute.levels=attribute_list,
constraints=constraint_list, level.probs=marginal_weights)
baselines <- list()
baselines$Education <- "High school graduate"
baselines$Background <- "US-born legal grandparents"
baselines$Profession <- "Teacher"
baselines$Political <- "Not political"
baselines$Religion <- "Not religious"
#### AMCE of Perceptions of Americanness (Amerian rating) ----
####NOTE: This model is computationally intensive and may not run on all computers
ratingamce_r <- cjoint::amce(rating_r ~ Faces + Background + Identifies +
Language + Education + Profession + Religion
+ Political + Spouse, data=df, cluster=TRUE,
respondent.id="caseid", design=conjointdesign,
weights="weight", baselines=baselines,
na.ignore=TRUE)
### Creating a table to visualize the results
sum_amce_r <- summary(ratingamce_r)
amce_r <-sum_amce_r$amce
amcer_table <- amce_r[c(2:6)]
tab_df(amcer_table)| Level | Estimate | Std..Err | z.value | Pr...z.. |
| Legal immigrant | -0.04 | 0.01 | -2.96 | 0.00 |
| Undoc. immigrant | -0.21 | 0.02 | -13.67 | 0.00 |
| Naturalized citizen | -0.02 | 0.01 | -1.35 | 0.18 |
| US-born legal parents | 0.02 | 0.01 | 2.19 | 0.03 |
| US-born undoc. parents | -0.05 | 0.01 | -4.10 | 0.00 |
| US-born undoc. grandparents | -0.02 | 0.01 | -2.46 | 0.01 |
| High school dropout | -0.00 | 0.01 | -0.40 | 0.69 |
| Bachelor's degree | 0.01 | 0.01 | 0.95 | 0.34 |
| Advanced/graduate degree | 0.01 | 0.01 | 0.67 | 0.51 |
| White woman | -0.01 | 0.01 | -0.91 | 0.36 |
| Light man | -0.00 | 0.01 | -0.22 | 0.83 |
| Light woman | -0.01 | 0.01 | -0.82 | 0.41 |
| Medium man | -0.01 | 0.01 | -0.79 | 0.43 |
| Medium woman | -0.01 | 0.01 | -0.77 | 0.44 |
| Brown man | -0.03 | 0.01 | -2.41 | 0.02 |
| Brown woman | -0.01 | 0.01 | -1.09 | 0.28 |
| Black man | -0.00 | 0.01 | -0.34 | 0.74 |
| Black woman | 0.01 | 0.01 | 0.62 | 0.53 |
| Latino | -0.02 | 0.01 | -2.03 | 0.04 |
| Hispanic | -0.02 | 0.01 | -2.57 | 0.01 |
| Mexican-American | -0.02 | 0.01 | -1.78 | 0.07 |
| Mexican | -0.03 | 0.01 | -2.80 | 0.01 |
| White | -0.01 | 0.01 | -0.98 | 0.33 |
| Black | 0.00 | 0.01 | 0.15 | 0.88 |
| Bilingual | 0.01 | 0.01 | 0.56 | 0.58 |
| Limited English | -0.02 | 0.01 | -2.03 | 0.04 |
| No English | -0.04 | 0.04 | -1.01 | 0.31 |
| Democrat | 0.01 | 0.01 | 1.17 | 0.24 |
| Republican | 0.00 | 0.01 | 0.46 | 0.65 |
| Independent | 0.01 | 0.01 | 2.15 | 0.03 |
| Waiter | -0.00 | 0.02 | -0.12 | 0.91 |
| Janitor | 0.01 | 0.01 | 0.58 | 0.56 |
| Sales manager | 0.00 | 0.01 | 0.15 | 0.88 |
| IT professional | 0.01 | 0.01 | 0.80 | 0.43 |
| Doctor | 0.01 | 0.02 | 0.35 | 0.73 |
| Small business owner | 0.02 | 0.01 | 1.59 | 0.11 |
| Presbyterian | 0.00 | 0.01 | 0.07 | 0.95 |
| Evangelical | -0.00 | 0.01 | -0.38 | 0.70 |
| Catholic | 0.00 | 0.01 | 0.61 | 0.54 |
| S. is American | 0.00 | 0.01 | 0.71 | 0.48 |
| S. is Black | -0.01 | 0.01 | -0.68 | 0.49 |
| S. is White | -0.01 | 0.01 | -0.75 | 0.45 |
| S. is same as profile | -0.00 | 0.01 | -0.69 | 0.49 |
1.6 Visualizations
Now let’s visualize the results on a figure based on the plot method. In order to do so, we will use a a combination of the method plot, implemented in cjoint with the ggplot in order to make the graph more presentable.
Show Answer
# Figure Limits
lim <- c(-.25, .05)
# Feature order
c_order <- c("Faces", "Identifies", "Religion", "Background", "Language",
"Spouse", "Education", "Profession", "Political")
figure_1 <- plot(ratingamce_r, main = "Average Marginal Component Effect of Perceived Americanness Among All Respondents",
xlab = "Change: American Rating (0 'not at all American' - 1 'very much American')", group.order= c_order,
colors="black",
xlim=lim,
point.size=0.4,
dodge.size=1,
plot.theme = (ggplot2::theme_bw() + theme(legend.position="none", plot.title= element_text(size=10)))) 
1.7 Extra features
The cregg package also allow us to perform the conjoint in a comparative way, seeing how different groups within the sample using the function mm_diffs. We again need to specify the design and add the argument by, which will specify the respondent groups that we are comparing.
Show Answer
###Marginal mean differences in perceptions of Americanness by race model----
american_diff <- cregg::mm_diffs(na.omit(df),
rating_r ~ Faces + Background
+ Education + Profession
+ Language + Political
+ Religion + Spouse
+ Identifies,
by = ~ race,
id= ~caseid,
weights = ~weight,
feature_order = c_order)
american_diff_table <- american_diff[c(5:9)]
tab_df(american_diff_table)| level | estimate | std.error | z | p |
| White man | 0.05 | 0.02 | 2.38 | 0.02 |
| White woman | 0.05 | 0.02 | 2.72 | 0.01 |
| Light man | 0.02 | 0.02 | 0.96 | 0.33 |
| Light woman | 0.05 | 0.02 | 2.65 | 0.01 |
| Medium man | 0.05 | 0.02 | 2.73 | 0.01 |
| Medium woman | 0.06 | 0.02 | 3.31 | 0.00 |
| Brown man | 0.04 | 0.02 | 1.90 | 0.06 |
| Brown woman | 0.03 | 0.02 | 1.36 | 0.17 |
| Black man | 0.03 | 0.02 | 1.86 | 0.06 |
| Black woman | 0.02 | 0.02 | 1.21 | 0.23 |
| American | 0.04 | 0.02 | 2.28 | 0.02 |
| Latino | 0.05 | 0.02 | 2.99 | 0.00 |
| Hispanic | 0.03 | 0.02 | 2.06 | 0.04 |
| Mexican-American | 0.03 | 0.02 | 2.10 | 0.04 |
| Mexican | 0.02 | 0.02 | 1.53 | 0.13 |
| White | 0.08 | 0.02 | 4.52 | 0.00 |
| Black | 0.03 | 0.02 | 1.26 | 0.21 |
| Presbyterian | 0.05 | 0.01 | 3.34 | 0.00 |
| Evangelical | 0.03 | 0.01 | 2.34 | 0.02 |
| Catholic | 0.03 | 0.01 | 2.12 | 0.03 |
| Not religious | 0.05 | 0.01 | 3.32 | 0.00 |
| Legal immigrant | 0.07 | 0.02 | 3.92 | 0.00 |
| Undoc. immigrant | -0.07 | 0.02 | -3.22 | 0.00 |
| Naturalized citizen | 0.07 | 0.02 | 3.80 | 0.00 |
| US-born legal parents | 0.07 | 0.01 | 4.50 | 0.00 |
| US-born undoc. parents | 0.00 | 0.02 | 0.14 | 0.89 |
| US-born legal grandparents | 0.07 | 0.02 | 4.47 | 0.00 |
| US-born undoc. grandparents | 0.04 | 0.02 | 2.73 | 0.01 |
| Only English | 0.06 | 0.01 | 3.93 | 0.00 |
| Bilingual | 0.04 | 0.01 | 3.55 | 0.00 |
| Limited English | 0.03 | 0.01 | 1.94 | 0.05 |
| No English | 0.03 | 0.02 | 1.10 | 0.27 |
| Not married | 0.05 | 0.01 | 3.48 | 0.00 |
| S. is American | 0.03 | 0.01 | 1.96 | 0.05 |
| S. is Black | 0.04 | 0.02 | 2.31 | 0.02 |
| S. is White | 0.05 | 0.02 | 3.25 | 0.00 |
| S. is same as profile | 0.03 | 0.01 | 1.83 | 0.07 |
| High school dropout | 0.03 | 0.02 | 1.86 | 0.06 |
| High school graduate | 0.04 | 0.02 | 2.44 | 0.01 |
| Bachelor's degree | 0.04 | 0.01 | 3.04 | 0.00 |
| Advanced/graduate degree | 0.05 | 0.01 | 3.36 | 0.00 |
| Waiter | 0.02 | 0.02 | 1.30 | 0.19 |
| Janitor | 0.05 | 0.02 | 2.89 | 0.00 |
| Sales manager | 0.05 | 0.02 | 2.88 | 0.00 |
| Teacher | 0.04 | 0.02 | 1.90 | 0.06 |
| IT professional | 0.05 | 0.02 | 2.37 | 0.02 |
| Doctor | 0.08 | 0.03 | 3.02 | 0.00 |
| Small business owner | 0.04 | 0.01 | 2.48 | 0.01 |
| Democrat | 0.03 | 0.01 | 1.95 | 0.05 |
| Republican | 0.05 | 0.01 | 3.59 | 0.00 |
| Independent | 0.04 | 0.01 | 2.56 | 0.01 |
| Not political | 0.04 | 0.01 | 2.88 | 0.00 |
To end, let`s visualize those differences.
Show Answer
american_diff_p <- plot(american_diff)
figure_2 <- american_diff_p + ggplot2::theme_bw() +
theme(legend.position="none", plot.title= element_text(size=10)) +
scale_color_manual(values = rep("black", 9)) +
geom_point(size = 2) +
ggplot2::labs(x = "Estimated Difference Between White and Black Respondents",
title = "Marginal Mean Differences in Perceived Americanness Between White and Black Respondents")
figure_2