In jsks/seqR: Ordinal Sequence Analysis

The seqR package provides a set of functions for ordinal sequence analysis with a focus on three methods:

1. Contigency tables illustrating observed pairs between all the different combinations of a given set of variables.

2. Percentage tables summarizing how often a given variable X is observed to be larger than Y, and vice versa.

3. Bubble plots for visualizing transitions between paired observations for two variables.

This vignette will provide an example analysis using a subset of the Varieties of Democracy dataset.

---

Warning: seqR currently assumes that your data is indexed at 0.

---

Data: vdem

The example V-Dem dataset (seqR::vdem) included in the package contains ordinalized versions of the Political Civil Liberties Index and the eleven related constituent lower-level indicators for the country Ghana from 1902-2016 (for full documentation, ?vdem). The data was taken as is with two exceptions:

1. The freedom of expression index (e_v2x_clpol_5C) was rescaled from 0-1 to 0-4.

2. The number of states for the variable v2meslfcen_ord was adjusted so that all variables (including our index) have 5 possible states (or values). This was done by the following function where V represents our vector of values for v2meslfcen_ord with a transformation from 4 states to 5.

```{R eval = F} function(V) round(((5 - 1) / (4 - 1)) * V

## Contingency Table
If we want to quickly create a contingency matrix showing the observed pairs between all of the different combinations of our variables we can use the function `ql_matrix`. 

The `S3` object returned is simply a matrix with the `ql_mat` inherited class. We can summarize our results and even decompose the table to show the individual relationships for a target variable.

```{R eval = F}
indicators <- setdiff(colnames(vdem), c("country_name", "year"))
full.matrix <- ql_matrix(vdem[, indicators], na.rm = T)

# Print a summarized table of our contingency matrix
summary(full.matrix)

# Convert to a list of matrices for each variable
ll.matrix <- as.list(full.matrix)

Isolating Increasing/Decreasing Sequences

The package includes the utility function findMovement, which, given a vector, returns the intervals of subsequences based on movement --- additional options, such as buffering and lower/upper thresholds can be explored in the function documentation, ?findMovement. At the risk of stating the obvious, it should be noted that order matters when passing a vector to findMovement.

For example, if we want to only look at the country-years where our civil liberties index is increasing:

```{R eval = F}

Return the intervals of increasing subsequences

v <- findMovement(vdem$e_v2x_clpol_5C, direction = "up")

Remove the rows from vdem where v != NA (ie, where clpol is decreasing)

vdem_inc <- vdem[!is.na(v), ] full.matrix2 <- ql_matrix(vdem_inc[, indicators], na.rm = T)

## Percentage tables
A percentage table indicates the percentage of the overall observations for each combinations of variables where one is larger than the other and vice versa. The `ptable` function will calculate a percentage table for all variables in a given data frame.


```{R eval = F}
ptable(vdem[, !colnames(vdem) %in% c("country_name", "year")])

To construct a ptable of only transitions we can use the utility function collapse to reduce our data frame. Useful if we're concerned about unchanging observations biasing our results.

```{R eval = F} vdem_ordered <- vdem[order(vdem$country_name, vdem$year), ] vdem_reduced <- collapse(vdem_ordered[, !colnames(vdem) %in% c("country_name", "year")])

nrow(vdem)

We've shrunk!

nrow(vdem_reduced)

ptable(vdem_reduced)

## Bubble Plots
Finally, we can create directed adjacency graphs to visualize the transitions between observed pairs for two variables using the `bubbleplot` function.

Alternatively, we can construct an adjacency matrix first using `adj_mat`. This is important if we have, for example, a list of sequences split by a grouping variable for which we would like to separately create adjacency matrices that we add together before plotting. The object returned can then be plotted using the normal `plot` function.

```{R eval = F}
bubbleplot(v2x_clpol_5C, v2clacfree, data = vdem, xlab = "clpol", ylab = "clacfree")

# Alternative method
m <- adj_matrix(vdem$e_v2x_clpol_5C, vdem$v2clacfree)
plot(m, xlab = "clpol", ylab = "clacfree")

jsks/seqR documentation built on May 9, 2019, 12:48 p.m.