hrs.Rmd

# Health and Retirement Study (HRS) {-}

[![License: GPL v3](https://img.shields.io/badge/License-GPLv3-blue.svg)](https://www.gnu.org/licenses/gpl-3.0) <img src='https://img.shields.io/badge/Tested%20Locally-Windows%20Laptop-brightgreen' alt='Local Testing Badge'>

This detailed longitudinal study of the elderly in the United States allows for findings such as, "Among community residents aged 55-64 years old in 1998, what share lived in nursing homes by 2020?"

* Many tables from different timepoints, most with one row per sampled respondent and linkable.

* A complex sample survey designed to generalize to Americans aged 50+ at each interview point.

* Released biennially since 1992.

* Administered by the [University of Michigan's Institute for Social Research](http://isr.umich.edu/) with data management by the [RAND Corporation](http://www.rand.org/) and cross-national harmonization by the [University of Southern California](https://g2aging.org/). Funded by the [National Institute on Aging](https://www.nia.nih.gov/) and the [Social Security Administration](https://www.ssa.gov/).

---

## Recommended Reading {-}

Four Example Strengths & Limitations:

✔️ [Multiple cohorts allow extended tracking of older individuals across time](https://hrs.isr.umich.edu/sites/default/files/biblio/OverviewofHRSPublicData_0.pdf#page=32)

✔️ [Linkable to Medicare, Medicaid, SSA records, also to genetic and biomarker data](https://hrs.isr.umich.edu/data-products/restricted-data/available-products)

❌ [Sample size may prevent analysis of smaller populations or rare events](https://hrs.isr.umich.edu/documentation/survey-design/response-rates)

❌ [Attritors may differ in many ways from the general population](https://dx.doi.org/10.7302/24703)

<br>

Three Example Findings:

1. [Among individuals aged 50+ between 2003 and 2016, 80% of those who experienced a long-term care hospital stay subsequently died or suffered a severe impairment within 30 months](https://doi.org/10.1001/jamanetworkopen.2024.13309).

2. [Wealth loss of 75%+ was negatively associated with subsequent cognitive function over 2012-2016](https://doi.org/10.1016/S2666-7568(23)00113-7).

3. [The total monetary cost of dementia in 2010 was between $157 billion and $215 billion](https://doi.org/10.1056/nejmsa1204629).

<br>

Two Methodology Documents:

> [Getting Started with the Health and Retirement Study](https://hrsonline.isr.umich.edu/sitedocs/dmgt/IntroUserGuide.pdf)

> [RAND HRS Longitudinal File 2020 (V1) Documentation](https://www.rand.org/content/dam/rand/www/external/labor/aging/dataprod/randhrs1992_2020v1.pdf)

<br>

One Haiku:

```{r}
# sankey diagram
# comes alive at fifty five
# till death? you respond
```

---

## Download, Import, Preparation {-}

1. Register at the HRS Data Portal at https://hrsdata.isr.umich.edu/user/register.

2. Choose `RAND HRS Longitudinal File 2020 Latest release: Mar 2023 (V1)`.

3. Download the STATA dataset `randhrs1992_2020v1_STATA.zip` dated 04/05/2023:

```{r eval = FALSE , results = "hide" }
library(haven)

hrs_fn <- file.path( path.expand( "~" ) , "randhrs1992_2020v1.dta" )

hrs_tbl <- read_dta( hrs_fn )

hrs_df <- data.frame( hrs_tbl )

names( hrs_df ) <- tolower( names( hrs_df ) )
```

### Save Locally \ {-}

Save the object at any point:

```{r eval = FALSE , results = "hide" }
# hrs_fn <- file.path( path.expand( "~" ) , "HRS" , "this_file.rds" )
# saveRDS( hrs_df , file = hrs_fn , compress = FALSE )
```

Load the same object:

```{r eval = FALSE , results = "hide" }
# hrs_df <- readRDS( hrs_fn )
```

### Survey Design Definition {-}
Construct a complex sample survey design:

This design generalizes to residents of the United States that were living in the community in 1996 (wave 3) and also still alive (and participating in the survey) as of 2020 (wave 15):
```{r eval = FALSE , results = "hide" }
library(survey)

hrs_design <- 
	svydesign(
		id = ~ raehsamp ,
		strata = ~ raestrat ,
		weights = ~ r3wtresp , 
		nest = TRUE ,
		data = subset( hrs_df , r3wtresp > 0 & inw15 == 1 )
	)
```

### Variable Recoding {-}

Add new columns to the data set:
```{r eval = FALSE , results = "hide" }
hrs_design <- 
	update( 
		hrs_design , 

		one = 1 ,
		
		working_in_1996 = r3work ,

		working_in_2020 = r15work ,

		marital_stat_1996 =
			factor( r3mstat , levels = 1:8 , labels =
				c( "Married" , "Married, spouse absent" ,
				"Partnered" , "Separated" , "Divorced" ,
				"Separated/divorced" , "Widowed" ,
				"Never married" ) ) ,
				
		marital_stat_2020 =
			factor( r15mstat , levels = 1:8 , labels =
				c( "Married" , "Married, spouse absent" ,
				"Partnered" , "Separated" , "Divorced" ,
				"Separated/divorced" , "Widowed" ,
				"Never married" ) )
	)
```

---

## Analysis Examples with the `survey` library \ {-}

### Unweighted Counts {-}

Count the unweighted number of records in the survey sample, overall and by groups:
```{r eval = FALSE , results = "hide" }
sum( weights( hrs_design , "sampling" ) != 0 )

svyby( ~ one , ~ marital_stat_1996 , hrs_design , unwtd.count )
```

### Weighted Counts {-}
Count the weighted size of the generalizable population, overall and by groups:
```{r eval = FALSE , results = "hide" }
svytotal( ~ one , hrs_design )

svyby( ~ one , ~ marital_stat_1996 , hrs_design , svytotal )
```

### Descriptive Statistics {-}

Calculate the mean (average) of a linear variable, overall and by groups:
```{r eval = FALSE , results = "hide" }
svymean( ~ h15ahous , hrs_design , na.rm = TRUE )

svyby( ~ h15ahous , ~ marital_stat_1996 , hrs_design , svymean , na.rm = TRUE )
```

Calculate the distribution of a categorical variable, overall and by groups:
```{r eval = FALSE , results = "hide" }
svymean( ~ marital_stat_2020 , hrs_design , na.rm = TRUE )

svyby( ~ marital_stat_2020 , ~ marital_stat_1996 , hrs_design , svymean , na.rm = TRUE )
```

Calculate the sum of a linear variable, overall and by groups:
```{r eval = FALSE , results = "hide" }
svytotal( ~ h15ahous , hrs_design , na.rm = TRUE )

svyby( ~ h15ahous , ~ marital_stat_1996 , hrs_design , svytotal , na.rm = TRUE )
```

Calculate the weighted sum of a categorical variable, overall and by groups:
```{r eval = FALSE , results = "hide" }
svytotal( ~ marital_stat_2020 , hrs_design , na.rm = TRUE )

svyby( ~ marital_stat_2020 , ~ marital_stat_1996 , hrs_design , svytotal , na.rm = TRUE )
```

Calculate the median (50th percentile) of a linear variable, overall and by groups:
```{r eval = FALSE , results = "hide" }
svyquantile( ~ h15ahous , hrs_design , 0.5 , na.rm = TRUE )

svyby( 
	~ h15ahous , 
	~ marital_stat_1996 , 
	hrs_design , 
	svyquantile , 
	0.5 ,
	ci = TRUE , na.rm = TRUE
)
```

Estimate a ratio:
```{r eval = FALSE , results = "hide" }
svyratio( 
	numerator = ~ h4ahous , 
	denominator = ~ h15ahous , 
	hrs_design ,
	na.rm = TRUE
)
```

### Subsetting {-}

Restrict the survey design to :
```{r eval = FALSE , results = "hide" }
sub_hrs_design <- subset( hrs_design , working_in_1996 == 1 )
```
Calculate the mean (average) of this subset:
```{r eval = FALSE , results = "hide" }
svymean( ~ h15ahous , sub_hrs_design , na.rm = TRUE )
```

### Measures of Uncertainty {-}

Extract the coefficient, standard error, confidence interval, and coefficient of variation from any descriptive statistics function result, overall and by groups:
```{r eval = FALSE , results = "hide" }
this_result <- svymean( ~ h15ahous , hrs_design , na.rm = TRUE )

coef( this_result )
SE( this_result )
confint( this_result )
cv( this_result )

grouped_result <-
	svyby( 
		~ h15ahous , 
		~ marital_stat_1996 , 
		hrs_design , 
		svymean ,
		na.rm = TRUE 
	)
	
coef( grouped_result )
SE( grouped_result )
confint( grouped_result )
cv( grouped_result )
```

Calculate the degrees of freedom of any survey design object:
```{r eval = FALSE , results = "hide" }
degf( hrs_design )
```

Calculate the complex sample survey-adjusted variance of any statistic:
```{r eval = FALSE , results = "hide" }
svyvar( ~ h15ahous , hrs_design , na.rm = TRUE )
```

Include the complex sample design effect in the result for a specific statistic:
```{r eval = FALSE , results = "hide" }
# SRS without replacement
svymean( ~ h15ahous , hrs_design , na.rm = TRUE , deff = TRUE )

# SRS with replacement
svymean( ~ h15ahous , hrs_design , na.rm = TRUE , deff = "replace" )
```

Compute confidence intervals for proportions using methods that may be more accurate near 0 and 1. See `?svyciprop` for alternatives:
```{r eval = FALSE , results = "hide" }
svyciprop( ~ working_in_2020 , hrs_design ,
	method = "likelihood" , na.rm = TRUE )
```

### Regression Models and Tests of Association {-}

Perform a design-based t-test:
```{r eval = FALSE , results = "hide" }
svyttest( h15ahous ~ working_in_2020 , hrs_design )
```

Perform a chi-squared test of association for survey data:
```{r eval = FALSE , results = "hide" }
svychisq( 
	~ working_in_2020 + marital_stat_2020 , 
	hrs_design 
)
```

Perform a survey-weighted generalized linear model:
```{r eval = FALSE , results = "hide" }
glm_result <- 
	svyglm( 
		h15ahous ~ working_in_2020 + marital_stat_2020 , 
		hrs_design 
	)

summary( glm_result )
```

---

## Replication Example {-}

This example matches statistics and confidence intervals to four digits from the Gateway to Global Aging's [An Introduction to HRS, RAND HRS Longitudinal File, and Harmonized HRS](https://youtu.be/D409oUPXWUU?t=3201):

1. Navigate to `Contributed Projects` at https://hrsdata.isr.umich.edu/data-products/contributed-projects.

2. Choose `Gateway Harmonized HRS` Latest release: Aug 2023 Version D

3. Download the STATA dataset `H_HRS_d_stata.zip` dated 09/12/2023

```{r eval = FALSE , results = "hide" }
harmonized_hrs_fn <- file.path( path.expand( "~" ) , "H_HRS_d.dta" )

harmonized_hrs_tbl <- read_dta( harmonized_hrs_fn )

harmonized_hrs_df <- data.frame( harmonized_hrs_tbl )

names( harmonized_hrs_df ) <- tolower( names( harmonized_hrs_df ) )
```

Merge on cluster and strata variables from the RAND HRS Longitudinal file:

```{r eval = FALSE , results = "hide" }	
harmonized_hrs_rand_df <-
	merge(
		harmonized_hrs_df ,
		hrs_df[ c( 'hhid' , 'pn' , 'raestrat' , 'raehsamp' ) ] ,
		by = c( 'hhid' , 'pn' )
	)

stopifnot( nrow( harmonized_hrs_rand_df ) == nrow( hrs_df ) )
```

Limit the survey design to respondents answering at least two of the five different life satisfaction questions in the 2014 (wave 12) psychosocial leave-behind survey:

```{r eval = FALSE , results = "hide" }
h12sc_df <- subset( harmonized_hrs_rand_df , r12scwtresp > 0 & inw12sc == 1 )

r12sc_design <-
	svydesign(
		~ raehsamp ,
		strata = ~ raestrat ,
		data = h12sc_df ,
		weights = ~ r12scwtresp ,
		nest = TRUE
	)
```

Reproduce the coefficient, standard error, and confidence intervals presented at 53:20 of the tutorial:

```{r eval = FALSE , results = "hide" }
result <- svymean( ~ r12lsatsc , r12sc_design , na.rm = TRUE )

stopifnot( round( coef( result ) , 4 ) == 4.9822 )
stopifnot( round( SE( result ) , 4 ) == 0.0226 )
stopifnot( round( confint( result , df = degf( r12sc_design ) ) , 4 ) == c( 4.9369 , 5.0276 ) )
```

---

## Analysis Examples with `srvyr` \ {-}

The R `srvyr` library calculates summary statistics from survey data, such as the mean, total or quantile using [dplyr](https://github.com/tidyverse/dplyr/)-like syntax. [srvyr](https://github.com/gergness/srvyr) allows for the use of many verbs, such as `summarize`, `group_by`, and `mutate`, the convenience of pipe-able functions, the `tidyverse` style of non-standard evaluation and more consistent return types than the `survey` package. [This vignette](https://cran.r-project.org/web/packages/srvyr/vignettes/srvyr-vs-survey.html) details the available features. As a starting point for HRS users, this code replicates previously-presented examples:

```{r eval = FALSE , results = "hide" }
library(srvyr)
hrs_srvyr_design <- as_survey( hrs_design )
```
Calculate the mean (average) of a linear variable, overall and by groups:
```{r eval = FALSE , results = "hide" }
hrs_srvyr_design %>%
	summarize( mean = survey_mean( h15ahous , na.rm = TRUE ) )

hrs_srvyr_design %>%
	group_by( marital_stat_1996 ) %>%
	summarize( mean = survey_mean( h15ahous , na.rm = TRUE ) )
```