Example re-tidying

library(dplyr)
library(reshape2)
library(gapminder)

# find countries whose life expectancy over time correlates
cors <- gapminder %>%
  acast(year ~ country, value.var = "lifeExp") %>%
  cor() %>%
  melt(varnames = c("country1", "country2"), value.name = "correlation") %>%
  tbl_df()

cors

## Source: local data frame [20,164 x 3]
## 
##       country1    country2 correlation
##         (fctr)      (fctr)       (dbl)
## 1  Afghanistan Afghanistan   1.0000000
## 2      Albania Afghanistan   0.9656953
## 3      Algeria Afghanistan   0.9868220
## 4       Angola Afghanistan   0.9855294
## 5    Argentina Afghanistan   0.9705203
## 6    Australia Afghanistan   0.9393751
## 7      Austria Afghanistan   0.9557228
## 8      Bahrain Afghanistan   0.9956190
## 9   Bangladesh Afghanistan   0.9466653
## 10     Belgium Afghanistan   0.9626319
## ..         ...         ...         ...

This hopped from one tidy format to another, allowing easy manipulation:

# what countries are most correlated, don't count countries
# matched to themselves
cors %>%
  filter(country1 != country2) %>%
  arrange(desc(correlation))

## Source: local data frame [20,022 x 3]
## 
##              country1           country2 correlation
##                (fctr)             (fctr)       (dbl)
## 1          Mauritania          Indonesia   0.9996291
## 2           Indonesia         Mauritania   0.9996291
## 3             Senegal            Morocco   0.9995515
## 4             Morocco            Senegal   0.9995515
## 5  West Bank and Gaza       Saudi Arabia   0.9995156
## 6        Saudi Arabia West Bank and Gaza   0.9995156
## 7              France             Brazil   0.9994246
## 8              Brazil             France   0.9994246
## 9             Reunion            Bahrain   0.9993649
## 10            Bahrain            Reunion   0.9993649
## ..                ...                ...         ...

(I realize that since almost all countries are just going up linearly this isn’t a particularly interesting correlation metric, but there are many datasets where such between-group correlations are interesting).