step-01.02-spells.RMD

---
title: "Spells"
author: "Jesse Lecy"
output:
  html_document:
    theme: readable
    df_print: paged
    highlight: zenburn
    toc: true
    self_contained: false
    number_sections: false
    css: clean.css
---



```
output: word_document
```



# SETUP

```{r, warning=FALSE, message=FALSE}

library( foreign )
library( memisc )
library( knitr )
library( dplyr )
library( xtable )
library( stargazer )
library( broom )
library( pander )



```



# LOAD DATA

```{r, warning=FALSE }


dat <- readRDS( "./Data/CompleteHazardSpells.rds" )

# lapply( dat, class )
#                   
# head( dat, 25 ) %>% pander
```




# SPELLS


## Inspect Spells


```{r}

dat.sub <- dat[ ,c("ein","fisyr","age","FS_Totrev_adj","prof","JustNowProfessionalized") ]

head( dat.sub, 100 ) %>% pander

```




## Typical Spell Lengths

NOTE - do we want to experiment with imputing observations backwards for sensitivity analysis?

NOTE - if we don't impute spell lengths backwards, should we set age to first year in dataset?

```{r}

# count of spells by length

table( table( dat$ein ) )

# proportion that professionalize

sum( dat$prof ) / length(unique(dat$ein))

```


Number that never professionalize -  are right-censored because of end of study.

```{r}
sum( dat$prof == 0 & dat$age == 5 )

# 2815 that never professioalize
# 7744 orgs

2815 / 7744


```


Those that drop out before professionalization:

```{r}

do.prof <- tapply( dat$prof, dat$ein, sum )

these.do.not <- names(do.prof)[ do.prof == 0 ]

dat.not <- dat[ dat$ein %in% these.do.not , ]

nrow( dat.not )

table( tapply( dat.not$age, dat.not$ein, max ) )

2472 / 7744 # do not report (fail)

2815 / 7744 # right-censored (alive in 2003 but small)

2457 / 7744 # professionalize

table( tapply( dat.not$age, dat.not$ein, max ) ) /

  table( dat$age )


```



## Observation Accounting

```{r}

table( dat$AGE )

table( dat$fisyr )

table( dat$age, dat$fisyr ) %>% kable

table( dat$fisyr, dat$prof ) %>% kable


```