Comments for Lab 09 - Exploring data in the 2020 National Elections Study

Author

Your Group Members Names Here

Published

April 10, 2024

Overview

In this lab, we’ll explore data from the National Election Studies 2024 Pilot Study. You find an outcome of interest and a variable you think predicts interesting variation in that outcome. You’ll figure out what recoding you need to do, do that recoding and describe the data. I’ll do the same, so can some template code to compare your work to.

Everything we’ll do today is something we’ve done before and is also something you’ll likely have to do a version of in your final project.

  1. Set up your work space (5 minutes)

  2. Download and load data from the NES into R (10 minutes)

  3. Explore the codebook for the 2024 Pilot Study (5-10 minutes)

  4. Get a high level overview of the data to figure out what recoding needs to be done (5-10 minutes)

  5. Recode the data (15 minutes)

  6. Describe the data (15-20 minutes)

  7. Formulate a set of research questions (10 minutes)

  8. Save the data (5 minutes)

One of these 8 tasks will be randomly selected as the graded question for the lab.

set.seed(342024)
graded_question <- sample(1:8,size = 1)
paste("Question",graded_question,"is the graded question for this week")
[1] "Question 6 is the graded question for this week"

Goals

This week’s lab will give you practice:

  • Download and loading data from your own computers (Q1,Q2)

  • Exploring a codebook to find interesting relevant variables (Q3)

  • Exploring data to understand what needs to be recoded (Q4)

  • Recoding data in a clear and precise manner (Q5)

    • Demonstrate how to recode multiple variables using acros()

  • Describing, summarizing, and exploring data through tables and figures (Q6)

    • Using kable() and functions from the kableExtra package to format tables
    • Combining multiple figures into single plots using ggarrange()from the `ggpubr pacakge

  • Formulating research questions based on initial explorations of the data (Q7)

  • Saving data so you can use it later (Q8)

Workflow

Please render this .qmd file

As with every lab, you should:

  • Download the file
  • Save it in your course folder
  • Update the author: section of the YAML header to include the names of your group members in attendance.
  • Render the document
  • Open the html file in your browser (Easier to read)
  • Write your code in the chunks provided under each section
  • Comment out or delete any test code you do not need
  • Render the document again after completing a section or chunk (Error checking)
  • Upload the final lab to Canvas.

1 Set up your workspace

In the code chunk below, please add code to load the package DeclareDesign

1.1 Load Packages

# Libraries
library(tidyverse)
library(haven)
library(labelled)
library(kableExtra)
library(ggpubr)
library(DeclareDesign)

1.2 Set your working directory in R Studio

In R studio set your working directory to the folder where this lab is saved by clicking > Session > Set Working Directory > To Source File Location

After doing so uncomment getwd() Should print out something like

“~/Desktop/pols1600/labs/”

Depending on where your lab is saved

# In the Top Panel of RStudio Click
# Session > Set Working Directory > Source File Location

# Uncomment to Check Where Your File is Saved 
# getwd()
Note

If getwd() says something like ‘~/Downloads/’ click: “File > Save As” and save this lab in your course folder. Then close the version 09-lab.qmd that was opened from your Downloads folder and open the version of 09-lab.qmd that now exists in your course folder.

2 Download and load the NES data into R

Please click here to go to the download page for the National Election Studies 2024 Pilot Study

  • Click the Log-in Button

  • Create an account with the NES

![]((https://pols1600.paultesta.org/labs/images/09-nes2.png){width=30%}

  • Right click (Ctrl-click) on the DTA button. This will allow you to select Save Link As... Which should in turn allow you to chose where the file is saved. Save it in the same folder as 09-lab.qmd

  • Unzip anes_pilot_2024_dta_20240319.zip.
    • On a Mac this creates a new folder called anes_pilot_2024_dta_20240319 in your course folder which contains the following files:
    • anes_pilot_2024_dta_20240319.dta the data
    • anes_pilot_2024_userguidecodebook_20240319.pdf the codebook for the data
    • On a PC i think i it asks you where you want to unzip the files. Please unzip them directly into your course folder.

  • Load the data using some version of the following commands:
## IF .dta file is in subfolder where lab is:
# df <- read_dta("~/anes_pilot_2024_dta_20240319/anes_pilot_2024_20240319.dta")

## IF .dta file is in SAME folder as lab:
# df <- read_dta("anes_pilot_2024_20240319.dta")

## IF nothing works, fear not, you can load the data from the web as a backup
df <- read_dta(url("https://pols1600.paultesta.org/files/data/anes_pilot_2024_20240319.dta"))

3 Explore the codebook

Please open the file anes_pilot_2024_userguidecodebook_20240319.pdf.

It should be in same folder as the data.

Use Control+F for keywords to quickly navigate through the codebook looking for questions and variables that interest you.

3.1 Identify a potential outcome of interest

In this and next week’s lab, I’ll be exploring factors that explain variation in the following outcome variables:

A measure

  • vchoice_rematch “Vote Trump or Biden in 2024”

    • 1 = Donald Trump
    • 2 = Joe Biden
    • -7 = No Answer
    • -1 = Inapplicable

And five measures of political participation in the 2020 campaign

  • mobil_talk “2020 campaign - Talk to others about candidates”

  • mobil_online “2020 campaign - Participate in online rallies”

  • mobil_rally “2020 campaign - Attend in person rallies”

  • mobil_button “2020 campaign - Wear a button or campaign sticker”

  • mobil_work “2020 campaign - Any other work to support candidates”

    • 1 = Yes
    • 2 = No
    • -1 = Inapplicable

Please find a variable that describes some outcome of interest to you and fill in the following

  • outcome_variable_name Question topic
    • variable values

3.2 Identify at least one variable that might predict variation in your outcome.

From a quick skim, I’ve selected the following potential predictors, which I will recode below:

  • age Age
  • educ Education
  • faminc_new Income
  • race Race
  • pid7 7 point party identification

Please find at least one more predictor which you think might explain variation in your outcome of interest and fill in the following

  • predictor_variable_name Question topic
    • variable values

For example, perhaps you’re interested in differences by gender, or ideology, or social media use. See if you can find variables that measure these concepts.

You only needed to identify one, but you can choose to explore more if you want. Don’t choose 50, unless you really like recoding data.

4 Get a high level overview of the data

In this section, we’ll get practice quickly looking at variables to see what, if anything needs to be recoded.

I had you download the .dta instead of the .csv version of the data, because the .dta includes value labels for the data, which makes it easier to understand what a specific number corresponds to substantively.

4.1 Examine the distributions and values of your outcome variable

Please uncomment and run the code below

# Vote Choice
get_variable_labels(df$vchoice_rematch)
[1] "Vote Trump or Biden in 2024"
get_value_labels(df$vchoice_rematch)
                    No Answer inapplicable, legitimate skip 
                           -7                            -1 
                 Donald Trump                     Joe Biden 
                            1                             2 
                                                            
                            8                             9 
table(df$vchoice_rematch,useNA = "ifany")

 -7  -1   1   2 
 23 151 869 866 
# Acts of Participation
# All variables start with mobil_ prefix
df %>% select(starts_with("mobil")) %>% names()
 [1] "mobil_talk"             "mobil_online"           "mobil_rally"           
 [4] "mobil_button"           "mobil_work"             "mobil_talk_skp"        
 [7] "mobil_online_skp"       "mobil_rally_skp"        "mobil_button_skp"      
[10] "mobil_work_skp"         "mobil_talk_pg_timing"   "mobil_online_pg_timing"
[13] "mobil_rally_pg_timing"  "mobil_button_pg_timing" "mobil_work_pg_timing"  
# Political Talk
get_variable_labels(df$mobil_talk)
[1] "2020 campaign - Talk to others about candidates"
get_value_labels(df$mobil_talk)
                    No Answer inapplicable, legitimate skip 
                           -7                            -1 
                          Yes                            No 
                            1                             2 
                                                            
                            8                             9 
table(df$mobil_talk,useNA = "ifany")

  -7   -1    1    2 
   1  150  704 1054 
# Save the names all variables related to acts of participation in 2020
the_participation_vars <- df %>% select(starts_with("mobil")) %>% names()
# Only keep the variables that measure participation and not survey timing
the_participation_vars <- the_participation_vars[1:5]

From quickly looking at my outcome variables, I know that I will want to:

  • Recode vchoice_rematch to dv_vote_trump2024 which
    • equals 1 if vchoice_rematch == 1
    • equals 0 if vchoice_rematch == 2
    • equals NA if vchoice_rematch < 0
  • Recode variables that start with mobil_* to variables that start with polpart_* and:
    • equal 1 if mobil_* == 1
    • equal 0 if mobil_* == 2
    • equals NA if mobil_* < 0`
  • Create dv_participation* which is the sum of respondents’ five responses to the recoded polpart_* variables

4.2 Describe any recoding of your outcome variable

In the code chunk below, please repeat this process for the outcome variable you selected in the previous section:

# Get a HLO of your outcome variable
  • Recode YOUR OUTCOME VARIABLE HERE to NAME FOR RECODED VARIABLE
    • Describe what values need to be recoded in the new variable

4.3 Examine the distributions and values of your predictor variable

Now I’ll repeat this process for my potential predictor variables.

Please uncomment and run the code below

# Age
get_variable_labels(df$age)
[1] "Profile variable: Age"
get_value_labels(df$age)
not asked 
       -9 
summary(df$age)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  -9.00   33.00   51.00   48.43   63.00   94.00 
# Education
get_variable_labels(df$educ)
[1] "Profile variable: Education"
get_value_labels(df$educ)
    No HS credential High school graduate         Some college 
                   1                    2                    3 
       2-year degree        4-year degree            Post-grad 
                   4                    5                    6 
table(df$educ)

  1   2   3   4   5   6 
 84 584 381 210 426 224 
summary(df$educ)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  1.000   2.000   3.000   3.514   5.000   6.000 
# Income
get_variable_labels(df$faminc_new)
[1] "Profile variable: Family income"
get_value_labels(df$faminc_new)
                    No Answer inapplicable, legitimate skip 
                           -7                            -1 
            Less than $10,000             $10,000 - $19,999 
                            1                             2 
            $20,000 - $29,999             $30,000 - $39,999 
                            3                             4 
            $40,000 - $49,999             $50,000 - $59,999 
                            5                             6 
            $60,000 - $69,999             $70,000 - $79,999 
                            7                             8 
            $80,000 - $99,999           $100,000 - $119,999 
                            9                            10 
          $120,000 - $149,999           $150,000 - $199,999 
                           11                            12 
          $200,000 - $249,999           $250,000 - $349,999 
                           13                            14 
          $350,000 - $499,999              $500,000 or more 
                           15                            16 
            Prefer not to say                               
                           97                           998 
                              
                          999 
table(df$faminc_new)

 -7   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15  16  97 
 38 119 126 222 142 141 140 122 149 144 120 109  88  31  20  10   4 184 
summary(df$faminc_new)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  -7.00    3.00    7.00   14.89   10.00   97.00 
# Race
get_variable_labels(df$race)
[1] "Race"
get_value_labels(df$race)
                    No Answer inapplicable, legitimate skip 
                           -7                            -1 
                        White                         Black 
                            1                             2 
                     Hispanic                         Asian 
                            3                             4 
              Native American             Two or more races 
                            5                             6 
                        Other                Middle Eastern 
                            7                             8 
                                                            
                           98                            99 
summary(df$race)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  1.000   1.000   1.000   1.751   2.000   8.000 
table(df$race,useNA = "ifany")

   1    2    3    4    5    6    7    8 
1270  242  239   44   17   69   27    1 
# Partisanship
get_variable_labels(df$pid7)
[1] "Profile variable: 7 point party identification"
get_value_labels(df$pid7)
           Strong Democrat   Not very strong Democrat 
                         1                          2 
             Lean Democrat                Independent 
                         3                          4 
           Lean Republican Not very strong Republican 
                         5                          6 
         Strong Republican                   Not sure 
                         7                          8 
                Don't know 
                         9 
summary(df$pid7)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max.    NA's 
  1.000   2.000   4.000   4.023   6.000   8.000      58 
table(df$pid7)

  1   2   3   4   5   6   7   8 
389 228 173 295 172 184 342  68

After taking a quick look at each variable, I know that I’ll want to do the following recoding:

  • age to age1
    • recode -9s to NA
  • educ no recoding needed
    • create indicator has_college_degree which equals 1 is educ > 4 and 0 otherwise
  • faminc_new to income
    • recode -7 and 97 to NA
  • race to race_5cat
    • race == 1 ~ "White"
    • race == 2 ~ "Black"
    • race == 3 ~ "Hispanic"
    • race == 4 ~ "Asian"
    • T ~ "Other" (Collapse other racial categories)
  • race to is_* binary indicators:
    • is_white = 1 if race==1, 0 otherwise
  • pid7 to partyid
    • recode pid7 == 8 to 4 (Classify Don't Knows as Independents)
  • pid7 to is_*: binary indicators:
    • is_dem = 1 if partyid < 4, 0 otherwise
    • is_rep = 1 if partyid > 4, 0 otherwise
    • is_ind = 1 if partyid == 4, 0 otherwise

4.4 Describe any recoding of your predictor(s)

In the code chunk below, please repeat this process for the additional predictor(s) you selected in the previous section:

# Get a HLO of your outcome variable
  • Recode YOUR OUTCOME VARIABLE HERE to NAME FOR RECODED VARIABLE
    • Describe what values need to be recoded in the new variable

5 Recode the data

Now we’ve got a plan of action for how we need to recode the data.

5.1 Recode the outcome variables

Please uncomment and run the following code chunk:

df %>%
  # Recode 2024 Vote Choice
  mutate(
    dv_vote_trump2024 = case_when(
      vchoice_rematch == 1 ~ 1,
      vchoice_rematch == 2 ~ 0,
      T ~ NA
    )
  ) %>% 
  # Recode Individual Acts of Participation
  mutate(across(all_of(the_participation_vars), 
                \(x) case_when(
                  x == 1 ~ 1,
                  x == 2 ~ 0,
                  T ~ NA
                ),
                .names = "polpart_{.col}"
                )
                
                ) %>% 
  # Create Additive Measure of Participation
  mutate(
    dv_participation = rowSums(
      select(.,starts_with("polpart")),
      na.rm = T)
    
  ) -> df

Please recode your outcome of interest as needed

Remember to save the output of your recode back into the dataframe df

# Recode your outcome of interest

5.2 Check the recoding of your outcome

It’s a good habit to compare your recoded variables to their original values, to make sure your code did what you thought it did.

Please uncomment and run the following:

# Check recodes

table(
  recode = df$dv_vote_trump2024,
  original = df$vchoice_rematch, 
  useNA = "ifany"
  )
      original
recode  -7  -1   1   2
  0      0   0   0 866
  1      0   0 869   0
  <NA>  23 151   0   0
table(
  recode = df$polpart_mobil_button,
  original = df$mobil_button, 
  useNA = "ifany"
  )
      original
recode   -1    1    2
  0       0    0 1386
  1       0  373    0
  <NA>  150    0    0
table(
  total = df$dv_participation,
  item = df$polpart_mobil_button,
  useNA = "ifany"
  )
     item
total   0   1 <NA>
    0 884   0  149
    1 364  62    1
    2  89 109    0
    3  38  71    0
    4  11  83    0
    5   0  48    0

So everything looks in order. I could have probably checked the dv_participation variable against all of its constituent items, but based off comparing it to polpart_mobil_button everything looks in order since: - there are no cases where polpart_mobil_button is 1 but dv_participation is 0 - there are no cases where dv_participation is at it’s max but polpart_mobil_button is 0. - dv_participation has the correct theoretical range from 0 acts to 5 acts.

Now do the same for your outcome variable.

Please check your recoded outcome against its original values

# Check recodes

5.3 Recode your predictor variables

Again, I’ve provide some demonstration code to recode the predictors listed above.

Please uncomment and run the following

df %>% 
  mutate(
    # Age
    age = ifelse(age < 0, NA, age),
    # Education
    education = educ,
    educ_f = to_factor(educ), #Convert to Factor for Plotting
    is_college_grad = ifelse(educ > 4,1,0),
    # Income
    income = case_when(
      faminc_new < 0 ~ NA,
      faminc_new > 0 & faminc_new >16 ~ NA,
      T ~ faminc_new
    ),
    # Race
    race_5cat = case_when(
      race < 5 ~ to_factor(race),
      T ~ "Other"
    ) %>% factor(., levels = c("White","Black","Hispanic","Asian","Other")),
    is_white = ifelse(race == 1, 1, 0),
    is_black = ifelse(race == 2, 1, 0),
    is_hispanic = ifelse(race == 3, 1, 0),
    is_asian = ifelse(race == 4, 1, 0),
    is_other = ifelse(race == 5, 1, 0),
    # Partisanship
    partyid = case_when(
      pid7 == 8 ~ 4,
      T ~ pid7
    ),
    is_dem = ifelse(partyid  < 4, 1, 0),
    is_rep = ifelse(partyid  > 4, 1, 0),
    is_ind = ifelse(partyid  == 4, 1, 0),
  ) -> df

Please recode your additional predictor(s) as needed

# Recode your additional predictor(s)

It’s a good habit to check your all your recoding – particularly if you’re doing something like summing over multiple columns – but for this lab, we’ll live dangerously.

6 Describe the Data

Now let’s get some practice summarizing our data, presenting these summaries as tables and figures, and interpreting our results

6.1 Create a table of summary statistics

Please uncomment and run the code below which demonstrates how to produce a nicely formatted table of summary statistics

# Vector of numeric variables to summarize
the_vars <- c(
  "dv_vote_trump2024", 
  "dv_participation",
  "age","education","income",
  "is_white","is_black","is_hispanic","is_asian","is_other",
  "partyid","is_dem","is_rep","is_ind"
)

# Vector of nicely formatted labels for variables
the_labels <- c(
  "Vote for Trump in '24",
  "Acts of Participation in `20",
  "Age","Education", "Income",
  "White", "Black","Hispanic","Asian","Other",
  "Party ID", "Democrat","Republican","Independent"
)

df_summary <- df %>% 
  select(all_of(the_vars)) %>%
  rename_with(~the_labels) %>% 
  pivot_longer(
    cols = everything(),
    names_to = "Variable"
  ) %>% 
  mutate(
    Variable = factor(Variable, levels = the_labels)
  ) %>% 
  group_by(Variable) %>% 
  summarise(
    Min = min(value,na.rm = T),
    p25 = quantile(value, prob = .25,na.rm = T),
    Median = quantile(value, prob = .5,na.rm = T),
    Mean = mean(value, na.rm = T),
    p75 = quantile(value, prob = .75,na.rm = T),
    Max = max(value,na.rm = T),
    `N missing` = sum(is.na(value))
  )

# Look at results
df_summary
# A tibble: 14 × 8
   Variable             Min        p25 Median    Mean   p75 Max      `N missing`
   <fct>                <dbl+lb> <dbl>  <dbl>   <dbl> <dbl> <dbl+lb>       <int>
 1 Vote for Trump in '…  0           0      1 5.01e-1   1    1 [No …         174
 2 Acts of Participati…  0           0      0 9.25e-1   1    5 [4-y…           0
 3 Age                  18          33     51 4.94e+1  63.8 94                31
 4 Education             1 [No …     2      3 3.51e+0   5    6 [Pos…           0
 5 Income                1 [No …     3      6 6.43e+0   9   16 [$50…         222
 6 White                 0           0      1 6.65e-1   1    1 [No …           0
 7 Black                 0           0      0 1.27e-1   0    1 [No …           0
 8 Hispanic              0           0      0 1.25e-1   0    1 [No …           0
 9 Asian                 0           0      0 2.30e-2   0    1 [No …           0
10 Other                 0           0      0 8.91e-3   0    1 [No …           0
11 Party ID              1 [No …     2      4 3.88e+0   6    7 [$60…          58
12 Democrat              0           0      0 4.27e-1   1    1 [No …          58
13 Republican            0           0      0 3.77e-1   1    1 [No …          58
14 Independent           0           0      0 1.96e-1   0    1 [No …          58

We can then format df_summary as table using knitr() and styling options from the kableExtra package:

kable(df_summary,
      digits = 2) %>% 
  kable_styling() %>% 
  pack_rows("Outcomes", start_row = 1, end_row = 2) %>% 
  pack_rows("Demographic Predictors", 3, 10) %>% 
  pack_rows("Political Predictors", 11, 14)
Variable Min p25 Median Mean p75 Max N missing
Outcomes
Vote for Trump in '24 0 0 1 0.50 1.00 1 174
Acts of Participation in `20 0 0 0 0.93 1.00 5 0
Demographic Predictors
Age 18 33 51 49.37 63.75 94 31
Education 1 2 3 3.51 5.00 6 0
Income 1 3 6 6.43 9.00 16 222
White 0 0 1 0.67 1.00 1 0
Black 0 0 0 0.13 0.00 1 0
Hispanic 0 0 0 0.13 0.00 1 0
Asian 0 0 0 0.02 0.00 1 0
Other 0 0 0 0.01 0.00 1 0
Political Predictors
Party ID 1 2 4 3.88 6.00 7 58
Democrat 0 0 0 0.43 1.00 1 58
Republican 0 0 0 0.38 1.00 1 58
Independent 0 0 0 0.20 0.00 1 58

6.2 Modify the table to include your predictors

Using the two previous code chunks as a template, update the code so that the table includes your chosen outcome and predictors.

# Copy, paste, and update code from datasummary_me to include your variables

6.3 Interpret the results of your table.

Please write a few sentences that provide a substantive interpretation of your table of descriptive statistics

Your reader should come away with an understanding of the characteristics of the respondents to this sample.

The National Election Study’s 2024 Pilot Study contains responses from 1909 individuals2. The typical respondent in the data was just under 50 years old, with some college, with an income in the range of $50k-$59k. Approximately two-thirds of the sample identified as white, with 13 percent of respondents identifying as Black, 13 percent as Hispanic, 2 percent as Asian. Forty-three percent of respondents identified as Democrats, 38 percent as Republicans, and 20 percent as Independents. The respondents were evenly split in who they would vote for 2024, with 50 percent saying they would Vote for Donald Trump, and 50 percent saying Joe Biden. In the 2020 campaign, the average respondent reported engaging in about 1 act of political participation.

Note, that technically, 024 Pilot Study contains three types of respondents described by sample_type:

  • 1,500 respondents who are part of a weighted sample that generalizes to the U.S. population
  • 113 respondents who completed the survey, but were not included in the panel-matching procedure used to construct construct a representative sample
  • 296 respondents who didn’t complete the whole survey.

So, technically speaking3, if we wanted to draw inferences about the proportion of American’s planning to to vote for Trump or Biden in 2024, we should only look at the 1,500 respondents in the weighted sample, and we should calculate that proportion using the sampling weights provided by weights.

Survey weights are complicated things, but the basically idea is that each observation is the sample is representative of observations in the population. Some types of observations will be over-represented in our sample – these are given smaller weights – while others are under-represented – these are given greater weights.

Quick look at the highest and lowest survey weights seems to suggest to me that weighting procedure is giving more weight to respondents with lower levels of education, and income, and less weight to higher incomes and education levels.

df %>% filter(
 weight > quantile(weight, .99, na.rm=T)
 ) %>% 
  select(weight, gender, age, race_5cat, education, income, partyid)
# A tibble: 15 × 7
   weight gender       age race_5cat education                income    partyid 
    <dbl> <dbl+lbl>  <dbl> <fct>     <dbl+lbl>                <dbl+lbl> <dbl+lb>
 1   7.00 1 [Male]      23 Hispanic  3 [Some college]          4 [$30,…  3 [Lea…
 2   3.33 1 [Male]      31 Other     2 [High school graduate] NA         2 [Not…
 3   2.96 2 [Female]    33 Hispanic  1 [No HS credential]     NA         4 [Ind…
 4   3.36 1 [Male]      68 Black     3 [Some college]         NA         4 [Ind…
 5   2.94 1 [Male]      67 Hispanic  1 [No HS credential]      2 [$10,…  2 [Not…
 6   4.13 1 [Male]      22 Hispanic  3 [Some college]         11 [$120…  4 [Ind…
 7   3.64 2 [Female]    58 Black     5 [4-year degree]         6 [$50,… NA      
 8   3.30 1 [Male]      43 Hispanic  5 [4-year degree]         7 [$60,…  4 [Ind…
 9   3.51 1 [Male]      69 White     2 [High school graduate]  4 [$30,…  7 [Str…
10   3.32 1 [Male]      74 White     2 [High school graduate]  6 [$50,…  3 [Lea…
11   3.32 1 [Male]      69 White     2 [High school graduate]  3 [$20,…  4 [Ind…
12   3.51 1 [Male]      75 White     2 [High school graduate]  2 [$10,…  7 [Str…
13   2.93 1 [Male]      41 Black     4 [2-year degree]         4 [$30,…  2 [Not…
14   4.20 2 [Female]    63 Black     6 [Post-grad]             3 [$20,…  6 [Not…
15   3.32 1 [Male]      67 White     2 [High school graduate]  5 [$40,…  3 [Lea…
df %>% filter(
  weight < quantile(weight, .01, na.rm=T)
) %>% 
  select(weight, gender, age, race_5cat, education, income, partyid)
# A tibble: 12 × 7
   weight gender       age race_5cat education                income     partyid
    <dbl> <dbl+lbl>  <dbl> <fct>     <dbl+lbl>                <dbl+lbl>  <dbl+l>
 1  0.314 1 [Male]      77 Black     6 [Post-grad]            10 [$100,… 1 [Str…
 2  0.409 2 [Female]    29 Black     6 [Post-grad]            14 [$250,… 1 [Str…
 3  0.382 1 [Male]      23 Hispanic  5 [4-year degree]        10 [$100,… 5 [Lea…
 4  0.439 2 [Female]    26 Asian     5 [4-year degree]        12 [$150,… 2 [Not…
 5  0.421 2 [Female]    57 Hispanic  6 [Post-grad]             6 [$50,0… 5 [Lea…
 6  0.415 2 [Female]    52 Hispanic  6 [Post-grad]             6 [$50,0… 3 [Lea…
 7  0.455 1 [Male]      72 Other     6 [Post-grad]             7 [$60,0… 7 [Str…
 8  0.405 1 [Male]      32 Hispanic  5 [4-year degree]        11 [$120,… 4 [Ind…
 9  0.362 2 [Female]    33 Black     5 [4-year degree]         1 [Less … 1 [Str…
10  0.411 1 [Male]      42 Black     2 [High school graduate]  4 [$30,0… 4 [Ind…
11  0.362 2 [Female]    33 Black     5 [4-year degree]         1 [Less … 1 [Str…
12  0.346 1 [Male]      71 Black     6 [Post-grad]             2 [$10,0… 1 [Str…

Sampling weights are a can worms we won’t deal with in this class. Check out these resources on the survey and srvyr packages for how to incorporate survey weights into your analysis. In the code below we see that distributions of race and ethnicity are pretty similar in the weighted and full samples, but using survey weights seems to suggest that Biden’s support as actually higher than Trumps, although the confidence intervals (Next week!) for both overlap 0.5 suggesting the race is essentially tied.

# install.packages("survey")
# install.packages("srvyr")
library(survey)
library(srvyr)

# Format as survey_design object
df_s <- as_survey_design(df %>% filter(sample_type == 1),weight = weight)

# Calculate weighted totals and proportions
df_s %>% 
  group_by(race) %>% 
  summarise(
    total = survey_total(),
    proption = survey_mean()
  )
# A tibble: 7 × 5
  race                   total total_se proption proption_se
  <dbl+lbl>              <dbl>    <dbl>    <dbl>       <dbl>
1 1 [White]             1005.     21.6   0.670       0.0139 
2 2 [Black]              193.     15.7   0.129       0.0102 
3 3 [Hispanic]           189.     15.9   0.126       0.0102 
4 4 [Asian]               35.6     6.36  0.0237      0.00423
5 5 [Native American]     11.8     3.26  0.00789     0.00218
6 6 [Two or more races]   46.8     7.52  0.0312      0.00499
7 7 [Other]               17.8     5.53  0.0119      0.00367
# Compare to unweighted estimates
df %>%
  group_by(race) %>% 
  summarise(
    total = n()
  ) %>% 
  mutate(
    propotion = total/sum(total)
  )
# A tibble: 8 × 3
  race                  total propotion
  <dbl+lbl>             <int>     <dbl>
1 1 [White]              1270  0.665   
2 2 [Black]               242  0.127   
3 3 [Hispanic]            239  0.125   
4 4 [Asian]                44  0.0230  
5 5 [Native American]      17  0.00891 
6 6 [Two or more races]    69  0.0361  
7 7 [Other]                27  0.0141  
8 8 [Middle Eastern]        1  0.000524
# Calculated weighted proportions of support
df_s %>% 
  group_by(dv_vote_trump2024) %>%
  summarise(
    proportion = survey_mean()
  ) %>% select(dv_vote_trump2024,proportion, proportion_se)
# A tibble: 3 × 3
  dv_vote_trump2024 proportion proportion_se
              <dbl>      <dbl>         <dbl>
1                 0     0.510        0.0142 
2                 1     0.476        0.0142 
3                NA     0.0139       0.00363
# Compare to unweighted proportions just among those in weighted sample
df %>%
  filter(sample_type == 1) %>% 
  group_by(dv_vote_trump2024) %>% 
  summarise(
    total = n()
  ) %>% 
  mutate(
    propotion = total/sum(total)
  )
# A tibble: 3 × 3
  dv_vote_trump2024 total propotion
              <dbl> <int>     <dbl>
1                 0   755     0.503
2                 1   727     0.485
3                NA    18     0.012
# And overall
df %>%
  group_by(dv_vote_trump2024) %>% 
  summarise(
    total = n()
  ) %>% 
  mutate(
    propotion = total/sum(total)
  )
# A tibble: 3 × 3
  dv_vote_trump2024 total propotion
              <dbl> <int>     <dbl>
1                 0   866    0.454 
2                 1   869    0.455 
3                NA   174    0.0911
# Formally test whether trumps support is above 50 percent
df_s %>% 
svyttest((dv_vote_trump2024==1) - 0 ~ 0,design = .,na.rm=T)

    Design-based one-sample t-test

data:  (dv_vote_trump2024 == 1) - 0 ~ 0
t = 33.732, df = 1498, p-value < 2.2e-16
alternative hypothesis: true mean is not equal to 0
95 percent confidence interval:
 0.4548575 0.5110241
sample estimates:
     mean 
0.4829408

6.4 Visualize the Data

Please a choose a variable or variables whose distribution or relationship you think may be substantively interesting to the potential story you want to tell.

In the code below, I visualize:

  • the distribution of age by race using geom_boxplot()
  • average partisanship by race using stat_summary()
  • how partisanship changes with age by race using geom_smooth()

And combine these 3 plots into a single figure using two calls to the ggarrange() function from the ggpubr package

fig_age_race <- df %>% 
  ggplot(aes(race_5cat,age,
             col = race_5cat))+
  geom_boxplot()+
  labs(
    x = "Race",
    y = "Age",
    col = "Race",
    title = "Distribution of Age by Race"
  )+
  theme_minimal()

fig_pid_race <- df %>% 
  ggplot(aes(race_5cat,partyid,
             col = race_5cat))+
  stat_summary(position = position_dodge(width=.5))+
  labs(
    x = "Race",
    y = "Partisanship",
    col = "Race",
    title = "Average Partisanship by Race"
  )+
  theme_minimal()

fig_age_race_pid <- df %>% 
  mutate(
    Race = race_5cat
  ) %>% 
  ggplot(aes(age,partyid,
             col = race_5cat
             ))+
  geom_smooth(se = F) +
  geom_jitter(size=.5, alpha=.3) +
  labs(
    x = "age",
    y = "Partisanship",
    col = "Race",
    title = "Distribution of Partisanship by Age and Race in 2024 NES Pilot Study"
  )+
  theme_minimal()

fig_desc <- ggarrange(
  # Top Row, two columns
  ggarrange(
    fig_age_race, fig_pid_race,
    ncol =2,
    legend = "none"
    ), 
  # Bottom row, 1 column
  fig_age_race_pid, 
  nrow=2,
  common.legend = T,
  legend = "bottom",
  heights = c(1,1.5)
  )

fig_desc

From the figure, we see that whites in sample have the highest median age (55 years) followed by Blacks (46 years). Hispanic and Asian respondents have the youngest median age of (40 years). Whites are also tend to lean more Republican in their partisanship than other racial minority groups. This is particularly true for older whites in the sample. Interestingly, average partisanship stays roughly constant with age for Asians and Hispanics, but older Black respondents are more likely to identify as Democrats than younger Blacks, whose partisan identification is more independent.

Please produce your own figure and provide a similar interpretation of what it conveys.

7 Formulate a research question

Please take a moment, formulate a research question you might ask of your data for next week’s lab

Here’s an example of some questions I might ask of these data

Research Question: Age, Race, and Support For Trump

Broadly, I’m interested in how support for Trump in the 2024 election varies with age and race. On average, I expect that older voters will be more supportive of Trump, but suspect that this trend varies by race. I expect it will be particularly true among White voters, but less so among People of color. Finally, I wonder whether these specific relationships hold, once we control for variations in partisan identification which we know varies both by age and race.

Research Question: Education, Income, and Political Participation

What’s the relationship between income, education, and political?4 Past research suggests both are associated with higher rates of political participation. But I wish to know whether the effects of one varies conditional on the levels of the other. Are income education complements, amplifying each others’ association with participation? Or are they possibly substitutes, such that increases in education for the low income might reduce gaps in rates of participation compared to higher income individuals?

8 Save your data

Finally, save your data so you can load all the work you’ve done today and hit the ground running, next lab.

# save(df, file = "nes24.rda")

Take the Class Survey

Please take a few moments to complete the class survey for this week.){target=“_blank”} for this week.

Footnotes

  1. In general, I try not to recode the original variables, but instead create new columns, with different names. But, like this footnote, it seems overly verbose to create something like age_recoded, so I’ll break my general rule↩︎

  2. See note below↩︎

  3. Which is the best kind of speaking↩︎

  4. In general, I’m often ambivalent on this kind of “What’s the effect of x on y” phrasing. I think it’s fine for framing, but the real question I’m interested in is something about the interaction and relative effects of income and education on participation, hence the subsequent clarification of research question.↩︎