Introduction

In this section, we describe the basics of using the household grid. Specifically, we show how to use the household grid to:

Identify particular family members
Create family-member specific variables
Determine the relationships between non-cohort members within a family.

We use the following packages:

# Load Packages
library(tidyverse) # For data manipulation
library(haven) # For importing .dta files

Finding Mother of Cohort Members

To show how to perform 1 & 2, we use the example of finding natural mothers’ smoking status at the first sweep. We load just four variables from the Sweep 1 household grid: MCSID and APNUM00, which together uniquely identify an individual, and AHPSEX00 and AHCREL00, which contain information on the individual’s sex and their relationship to the household’s cohort member(s). AHCREL00 == 7 identifies natural parents and AHPSEX00 == 2 identifies females. Combining the two identifies natural mothers. Below, we use count() to show the different (observed) values for the sex and relationship variables. We also use the filter() function (which retains observations where the conditions are TRUE) to create a dataset containing the identifiers (MCSID and APNUM00) of natural mothers only; we will merge this with the smoking information shortly. add_count(MCSID) %>% filter(n == 1) is included as an interim step to ensure there is just one natural mother per family.¹

df_0y_hhgrid <- read_dta("0y/mcs1_hhgrid.dta") %>%
  select(MCSID, APNUM00, AHPSEX00, AHCREL00) # Retains the listed variables

df_0y_hhgrid %>%
  count(AHPSEX00) # Tabulates each sex; AHPSEX00 does not record the sex of cohort members

# A tibble: 4 × 2
  AHPSEX00                n
  <dbl+lbl>           <int>
1 -2 [Unknown]           55
2 -1 [Not applicable] 18734
3  1 [Male]           26438
4  2 [Female]         29567

df_0y_hhgrid %>%
  count(AHCREL00) # Tabulates each relationship to a cohort member

# A tibble: 16 × 2
   AHCREL00                                n
   <dbl+lbl>                           <int>
-9 [Refusal]                            5
-8 [Dont Know]                          1
7 [Natural parent]                 33812
8 [Adoptive parent]                    2
9 [Foster parent]                      3
10 [Step-parent/partner of parent]     50
11 [Natural brother/Natural sister] 13873
12 [Half-brother/Half-sister]        3486
13 [Step-brother/Step-sister]          16
14 [Adopted brother/Adopted sister]     8
15 [Foster brother/Foster sister]       9
17 [Grandparent]                     2164
18 [Nanny/au pair]                     20
19 [Other relative]                  2326
20 [Other non-relative]               233
96 [Self]                           18786

df_0y_mothers <- df_0y_hhgrid %>%
  filter(
    AHCREL00 == 7, # Keep natural parents...
    AHPSEX00 == 2 # ...who are female.
  ) %>%
  add_count(MCSID) %>% # Creates new variable (n) containing # of records with given MCSID
  filter(n == 1) %>% # Keep where only one recorded natural mother per family
  select(MCSID, APNUM00) # Keep identifier variables

Note, where a cohort member is part of a family (MCSID) with two or more cohort members, the cohort member will have been a multiple birth (i.e., twin or triplet), so familial relationships should apply to all cohort members in the family, which is why there is just one relationship ([A-G]HCREL00) variable per household grid file. This will change as the cohort members age, moving into separate residences and starting their own families.

Creating a Mother’s Smoking Variable

Now we have a dataset containing the IDs of natural mothers, we can load the smoking information from the Sweep 1 parent interview file (mcs1_parent_interview.dta). The smoking variable we use is called APSMUS0A and contains information on the tobacco product (if any) a parent consumes. We classify a parent as a smoker if they use any tobacco product (mutate(parent_smoker = case_when(...))).

df_0y_parent <- read_dta("0y/mcs1_parent_interview.dta") %>%
  select(MCSID, APNUM00, APSMUS0A) # Retains only the variables we need

df_0y_parent %>%
  count(APSMUS0A)

# A tibble: 9 × 2
  APSMUS0A                            n
  <dbl+lbl>                       <int>
-9 [Refusal]                        4
-8 [Don't Know]                     3
-1 [Not applicable]                10
1 [No, does not smoke]         21229
2 [Yes, cigarettes]             9003
3 [Yes, roll-ups]               1246
4 [Yes, cigars]                  217
5 [Yes, a pipe]                    6
95 [Yes, other tobacco product]    16

df_0y_smoking <- df_0y_parent %>%
  mutate(parent_smoker = case_when(APSMUS0A %in% 2:95 ~ 1, # If APSMUS0A is integer between 2 and 95, then 1
                            APSMUS0A == 1 ~ 0)) %>% # If APSMUS0A is 1, then 0
  select(MCSID, APNUM00, parent_smoker)

Now we can merge the two datasets together to ensure we only keep rows in df_0y_smoking that appear in df_0y_mothers. We use left_join() to do this, with df_0y_mothers as the dataset determining the outputted rows, so that we have one row per identified mother.² The result is a dataset with one row per family with an identified mother. We rename the parent_smoker variable to mother_smoker to clarify that it refers to the mother’s smoking status.

Below we also pipe this dataset into the tabyl() function (from janitor) to tabulate the number and proportions of mothers who smoke and those who do not.

# install.packages("janitor") # Uncomment if you need to install
library(janitor)
df_0y_mothers %>%
  left_join(df_0y_smoking, by = c("MCSID", "APNUM00")) %>%
  select(MCSID, mother_smoker = parent_smoker) %>%
  tabyl(mother_smoker)

 mother_smoker     n     percent valid_percent
             0 12883 0.695814205     0.6968304
             1  5605 0.302727518     0.3031696
            NA    27 0.001458277            NA

Determining Relationships between Non-Cohort Members

The household grids include another set of relationship variables besides [A-G]HCREL00. These vary in name slightly between sweeps: [A-D]HPREL[A-Z]0 in mcs[1-4]_hhgrid.dta, EPREL0[A-Z]00 in mcs5_hhgrid.dta, and [F-G]HPREL0[A-Z] in mcs[6-7]_hhgrid.dta. These variables can be used to identify the relationships between non-cohort member family members. Specifically, they record the person in the row’s (ego) relationship to the person denoted by the column (alt); the letter [A-Z] in the variable name corresponds to the alt’s [A-D]PNUM00. For instance, the variable AHPRELB0 denotes the relationship of the person in the row to the person in the same family with APNUM00 == 2. Below, we extract a small set of data from the Sweep 1 household grid to show this in action.

df_0y_hhgrid_prel <- read_dta("0y/mcs1_hhgrid.dta") %>%
  select(MCSID, APNUM00, matches("AHPREL[A-Z]0"))

df_0y_hhgrid_prel %>%
  filter(MCSID == "M10001N") %>% # To look at just one family
  select(APNUM00, AHPRELA0, AHPRELB0, AHPRELC0) # To look at first few relationship variables

# A tibble: 7 × 4
  APNUM00 AHPRELA0                  AHPRELB0                  AHPRELC0          
    <dbl> <dbl+lbl>                 <dbl+lbl>                 <dbl+lbl>         
     1 96 [Self]                  1 [Husband/Wife]          7 [Natural paren…
     2  1 [Husband/Wife]         96 [Self]                  7 [Natural paren…
     3  3 [Natural son/daughter]  3 [Natural son/daughter] 96 [Self]         
     4  3 [Natural son/daughter]  3 [Natural son/daughter] 11 [Natural broth…
     5  3 [Natural son/daughter]  3 [Natural son/daughter] 11 [Natural broth…
     6  3 [Natural son/daughter]  3 [Natural son/daughter] 11 [Natural broth…
   100  3 [Natural son/daughter]  3 [Natural son/daughter] 11 [Natural broth…

There are seven members in this family, one of whom is a cohort member (APNUM00 == 100). APNUM00’s 1 and 2 are the (natural) parents, and APNUM00’s 3-6 and 100 are the (natural) children. The relationship variables show that APNUM00’s 1 and 2 are married, and APNUM00’s 3-7 are siblings (AHPRELC0 == 11 [Natural brother/sister]) and biological offspring of APNUM00’s 1 and 2 (AHPREL[A-B]0 == 3 [Natural son/daughter]). Note the symmetry in the relationships. Where, APNUM00 == 1, AHPRELC0 == 7 [Natural Parent] and where APNUM00 == 3, AHPRELA0 == 3 [Natural son/daughter].

If we want to find the particular person occupying a specific relationship for an individual (e.g., we want to know the [A-G]PNUM00 of the person’s partner), we need to reshape the data into long-format with one row per ego-alt relationship within a family. For instance, if we want to find each person’s spouse (conditional on one being present), we can do the following:³

df_0y_hhgrid_prel %>%
  pivot_longer(cols = matches("AHPREL[A-Z]0"),
               names_to = "alt",
               values_to = "relationship") %>%
  mutate(APNUM00_alt = match(str_sub(alt, -2, -2), LETTERS)) %>% # Creates alt's PNUM00 by matching penultimate letter to position in alphabet
  filter(relationship == 1) %>% # Keep where husband or wife
  select(MCSID, APNUM00, partner_pnum = APNUM00_alt)

# A tibble: 23,616 × 3
   MCSID   APNUM00 partner_pnum
   <chr>     <dbl>        <int>
M10001N       1            2
M10001N       2            1
M10002P       1            2
M10002P       2            1
M10007U       1            2
M10007U       2            1
M10011Q       1            2
M10011Q       2            1
M10015U       1            2
M10015U       2            1
# ℹ 23,606 more rows

Coda

This only scratches the surface of what can be achieved with the household grid. The mcs[1-7]_hhgrid.dta files also contain information on cohort-member and family-member’s dates of birth, which can be used to, for example, identify the number of resident younger siblings, determine maternal and paternal age at birth, and so on.

Footnotes

Loading the .dta files into R with haven::read_dta() retains the dataset metadata, including variable names and labels, mainly by storing variables as labelled class objects. See the labelled package help files for more information on working with this metadata - for instance, converting labelled variables to standard R factor variables or replacing negative values (generally reserved in MCS data to indicate missingness) with R’s native NA value. ↩
left_join() takes as arguments two data frames and retains only the rows in the first data frame, regardless of whether there is a match with the second. See Combining Data Across Sweeps for more discussion of the *_join() functions. ↩
For more on reshaping data, see Reshaping Data from Long to Wide (or Wide to Long) for more discussion of the *_join() functions. ↩