Skip to contents

Calculate the local univariate join count

local_joincount_uni.Rd

The univariate local join count statistic is used to identify clusters of rarely occurring binary variables. The binary variable of interest should occur less than half of the time.

Usage

local_joincount_uni(
  fx,
  chosen,
  listw,
  alternative = "two.sided",
  nsim = 199,
  iseed = NULL,
  ties.method = "average",
  no_repeat_in_row=FALSE
)

Arguments

fx

a factor with two levels; use of an ordered factor is not well understood.

chosen

a scalar character containing the level of fx that should be considered the observed value (1).

listw

a listw object containing binary weights created, for example, with nbwlistw(nb, style = "B")

alternative

default "greater". One of "less" or "greater".

nsim

the number of conditional permutation simulations

iseed

default NULL, used to set the seed; the output will only be reproducible if the count of CPU cores across which computation is distributed is the same

ties.method

default "average" passed through to rank, can take values accepted by rank: c("average", "first", "last", "random", "max", "min"), see rank

no_repeat_in_row

default FALSE, if TRUE, sample conditionally in each row without replacements to avoid duplicate values, https://github.com/r-spatial/spdep/issues/124

Value

a data.frame with class “local-jc-uni” with six columns BB (observed BB - neighbour same as focus), Pr() (pseudo-p from punif rank), sim_rank (simulation rank with current ties.method), p_sim_pysal_ge (pseudo-p fromPySAL esda, greater than or equal to observed BB), p_sim_pysal_gt (pseudo-p fromPySAL esda, greater than observed BB), largereq (count of simulated values greater than or equal to observed BB) after folding, olarger (count of simulated values greater than observed BB) before folding, olargereq (count of simulated values greater than or equal to observed BB) before folding,and number of rows equal to the length of x.

Details

The local join count statistic requires a binary weights list which can be generated with nb2listw(nb, style = "B"). Additionally, ensure that the binary variable of interest is rarely occurring in no more than half of observations.

P-values are estimated using a conditional permutation approach. This creates a reference distribution from which the observed statistic is compared. For more see Geoda Glossary.

The pseudo-p-values returned by freestanding Geoda and PySAL esda correspond to "res_min$p_sim_pysal_ge" and ranked pseudo-p-values with tied.method="min" (equivalently "last"; "res_min$p_sim_pysal_gt" corresponds to ranked pseudo-p-values with tied.method="max" (equivalently "first").

References

Anselin, L., & Li, X. (2019). Operational Local Join Count Statistics for Cluster Detection. Journal of geographical systems, 21(2), 189–210. doi:10.1007/s10109-019-00299-x

Author

Josiah Parry josiah.parry@gmail.com

Examples

data(oldcol)
fx <- as.factor(ifelse(COL.OLD$CRIME < 35, "low-crime", "high-crime"))
listw <- nb2listw(COL.nb, style = "B")
set.seed(1)
res_min <- local_joincount_uni(fx, chosen = "high-crime", listw, nsim=999,
 alternative="two.sided", ties.method="min")
cor(res_min[,2], res_min$p_sim_pysal_ge, use="complete.obs")
#> [1] 1
res_max <- local_joincount_uni(fx, chosen = "high-crime", listw, nsim=999,
 alternative="two.sided", ties.method="max")
cor(res_max[,2], res_max$p_sim_pysal_gt, use="complete.obs")
#> [1] 1