cm_distance: Distance Matrix Between Codes

Description

Generate distance measures to ascertain a mean distance measure between codes.

Usage

cm_distance(
  dataframe,
  pvals = c(TRUE, FALSE),
  replications = 1000,
  parallel = TRUE,
  extended.output = TRUE,
  time.var = TRUE,
  code.var = "code",
  causal = FALSE,
  start.var = "start",
  end.var = "end",
  cores = detectCores()/2
)

Arguments

dataframe

A data frame from the cm_x2long family (cm_range2long; cm_df2long; cm_time2long).

pvals

A logical vector of length 1 or 2. If element 2 is blank element 1 will be recycled. If the first element is TRUE pvalues will be calculated for the combined (main) output for all repeated measures from simulated resampling of the data. If the second element is TRUE pvalues will be calculated for the individual (extended) repeated measures output from simulated resampling of the data. Default is to calculate pvalues for the main output but not for the extended output. This process involves multiple resampling of the data and is a time consuming process. It may take from a few minutes to days to calculate the pvalues depending on the number of all codes use, number of different codes and number of replications.

replications

An integer value for the number of replications used in resampling the data if any pvals is TRUE. It is recommended that this value be no lower than 1000. Failure to use enough replications may result in unreliable pvalues.

parallel

logical. If TRUE runs the cm_distance on multiple cores (if available). This will generally be effective with most data sets, given there are repeated measures, because of the large number of simulations. Default uses 1/2 of the available cores.

extended.output

logical. If TRUE the information on individual repeated measures is calculated in addition to the aggregated repeated measures results for the main output.

time.var

An optional variable to split the dataframe by (if you have data that is by various times this must be supplied).

code.var

The name of the code variable column. Defaults to "codes" as out putted by x2long family.

causal

logical. If TRUE measures the distance between x and y given that x must precede y. That is, only those $y_{i}$ that begin after the $x_{i}$ has begun will be considered, as it is assumed that x precedes y. If FALSE x is not assumed to precede y. The closest $y_{i}$ (either its beginning or end) is calculated to $x_{i}$ (either it's beginning or end).

start.var

The name of the start variable column. Defaults to "start" as out putted by x2long family.

end.var

The name of the end variable column. Defaults to "end" as out putted by x2long family.

cores

An integer value describing the number of cores to use if parallel = TRUE. Default is to use half of the available cores.

Value

An object of the class "cm_distance". This is a list with the following components:

pvals

A logical indication of whether pvalues were calculated

replications

Integer value of number of replications used

extended.output

An optional list of individual repeated measures information

main.output

A list of aggregated repeated measures information

adj.alpha

An adjusted alpha level (based on $α = .05$ ) for the estimated p-values using the upper end of the confidence interval around the p-values

Within the lists of extended.output and list of the main.output are the following items:

mean

A distance matrix of average distances between codes

A matrix of standard deviations of distances between codes

A matrix of counts of distances between codes

stan.mean

A matrix of standardized values of distances between codes. The closer a value is to zero the closer two codes relate.

pvalue

A n optional matrix of simulated pvalues associated with the mean distances

Warning

p-values are estimated and thus subject to error. More replications decreases the error. Use:

$p \pm (1.96 \cdot \sqrt{\frac{α (1 - α)}{n}})$

to adjust the confidence in the estimated p-values based on the number of replications.

Details

Note that row names are the first code and column names are the second comparison code. The values for Code A compared to Code B will not be the same as Code B compared to Code A. This is because, unlike a true distance measure, cm_distance's matrix is asymmetrical. cm_distance computes the distance by taking each span (start and end) for Code A and comparing it to the nearest start or end for Code B.

References

https://stats.stackexchange.com/a/22333/7482

Examples

Run this code

# NOT RUN {
foo <- list(
    AA = qcv(terms="02:03, 05"),
    BB = qcv(terms="1:2, 3:10"),
    CC = qcv(terms="1:9, 100:150")
)

foo2  <- list(
    AA = qcv(terms="40"),
    BB = qcv(terms="50:90"),
    CC = qcv(terms="60:90, 100:120, 150"),
    DD = qcv(terms="")
)

(dat <- cm_2long(foo, foo2, v.name = "time"))
plot(dat)
(out <- cm_distance(dat, replications=100))
names(out)
names(out$main.output)
out$main.output
out$extended.output
print(out, new.order = c(3, 2, 1))
print(out, new.order = 3:2)
#========================================
x <- list(
    transcript_time_span = qcv(00:00 - 1:12:00),
    A = qcv(terms = "2.40:3.00, 6.32:7.00, 9.00,
        10.00:11.00, 59.56"),
    B = qcv(terms = "3.01:3.02, 5.01,  19.00, 1.12.00:1.19.01"),
    C = qcv(terms = "2.40:3.00, 5.01, 6.32:7.00, 9.00, 17.01")
)
(dat <- cm_2long(x))
plot(dat)
(a <- cm_distance(dat, causal=TRUE, replications=100))

## Plotting as a network graph
datA <- list(
    A = qcv(terms="02:03, 05"),
    B = qcv(terms="1:2, 3:10, 45, 60, 200:206, 250, 289:299, 330"),
    C = qcv(terms="1:9, 47, 62, 100:150, 202, 260, 292:299, 332"),
    D = qcv(terms="10:20, 30, 38:44, 138:145"),
    E = qcv(terms="10:15, 32, 36:43, 132:140"),
    F = qcv(terms="1:2, 3:9, 10:15, 32, 36:43, 45, 60, 132:140, 250, 289:299"),
    G = qcv(terms="1:2, 3:9, 10:15, 32, 36:43, 45, 60, 132:140, 250, 289:299"),
    H = qcv(terms="20, 40, 60, 150, 190, 222, 255, 277"),
    I = qcv(terms="20, 40, 60, 150, 190, 222, 255, 277")
)

datB  <- list(
    A = qcv(terms="40"),
    B = qcv(terms="50:90, 110, 148, 177, 200:206, 250, 289:299"),
    C = qcv(terms="60:90, 100:120, 150, 201, 244, 292"),
    D = qcv(terms="10:20, 30, 38:44, 138:145"),
    E = qcv(terms="10:15, 32, 36:43, 132:140"),
    F = qcv(terms="10:15, 32, 36:43, 132:140, 148, 177, 200:206, 250, 289:299"),
    G = qcv(terms="10:15, 32, 36:43, 132:140, 148, 177, 200:206, 250, 289:299"),
    I = qcv(terms="20, 40, 60, 150, 190, 222, 255, 277")  
)

(datC <- cm_2long(datA, datB, v.name = "time"))
plot(datC)
(out2 <- cm_distance(datC, replications=1250))

plot(out2)
plot(out2, label.cex=2, label.dist=TRUE, digits=5)
# }

Run the code above in your browser using DataLab

Last chance! 50% off unlimited learning