NHIS2009.Rmd

---
output: html_document
editor_options:
  chunk_output_type: console
---
# National Health Interview Survey

This reproduces the analyses in Table 1.1 of @AngristPischke2014.
which compares people with and without health insurance in the 2009 National Health Interview Survey (NHIS).

The code is derived from  [NHIS2009_hicompare.do](http://masteringmetrics.com/wp-content/uploads/2015/01/NHIS2009_hicompare.do).

Load the prerequisite packages.
```{r libs,message=FALSE}
library("tidyverse")
library("magrittr")
library("haven")
```

Load the data (originally from <http://masteringmetrics.com/wp-content/uploads/2015/01/Data.zip>), and adjust a few of the columns to account for differences in
how Stata and R store data.
```{r NHIS2009}
data("NHIS2009", package = "masteringmetrics")
```
Remove missing values.
```{r NHIS2009_remove_missing}
NHIS2009 <- NHIS2009 %>%
  filter(marradult, perweight != 0) %>%
  group_by(serial) %>%
  mutate(hi_hsb = mean(hi_hsb1, na.rm = TRUE)) %>%
  filter(!is.na(hi_hsb), !is.na(hi)) %>%
  mutate(female = sum(fml)) %>%
  filter(female == 1) %>%
  select(-female)
```

For the sample only include married adults between 26 and 59 in age, and remove single person households.
```{r NHIS2009_sample}
NHIS2009 <- NHIS2009 %>%
  filter(between(age, 26, 59),
         marradult, adltempl >= 1)

```
Keep only single family households.
```{r NHIS2009_single_family}
NHIS2009 <- NHIS2009 %>%
  group_by(serial) %>%
  filter(length(serial) > 1L) %>%
  ungroup()
```

Tables of wives and husbands by health insurance. status.
The weighting following the "analytic" weights in the original `.do` file which weights observations by `perweight` and normalizes the weights so that the sub-samples of males and females have the same number as the original sample.
```{r}
NHIS2009 %>%
  group_by(fml) %>%
  # normalize person weights to match number of observations in each
  # group
  mutate(perweight = perweight / sum(perweight) * n()) %>%
  group_by(fml, hi) %>%
  summarise(n_wt = sum(perweight)) %>%
  group_by(fml) %>%
  mutate(prop = n_wt / sum(n_wt))

```

Compare sample statistics of mean and women, with and without health insurance.
```{r NHIS2009_diff}
varlist <- c("hlth", "nwhite", "age", "yedu", "famsize", "empl", "inc")
NHIS2009_diff <- NHIS2009 %>%
  # rlang::set_attrs with NULL removes attributes from columns.
  # this avoids a warning from gather about differing attributes
  map_dfc(~ rlang::set_attrs(.x, NULL)) %>%
  select(fml, hi, one_of(varlist)) %>%
  gather(variable, value, -fml, -hi) %>%
  group_by(fml, hi, variable) %>%
  summarise(mean = mean(value, na.rm = TRUE), sd = sd(value, na.rm = TRUE)) %>%
  gather(stat, value, -fml, -hi, -variable) %>%
  unite(stat_hi, stat, hi) %>%
  spread(stat_hi, value) %>%
  mutate(diff = mean_1 - mean_0)
```

```{r NHIS2009_diff_table,results='asis'}
knitr::kable(NHIS2009_diff, digits = 3)
```

## References

-   <http://masteringmetrics.com/wp-content/uploads/2014/12/ReadMe_NHIS.txt>
-   <http://masteringmetrics.com/wp-content/uploads/2015/01/NHIS2009_hicompare.do>