ipd: Item Parameter Drift

In irt: Item Response Theory and Computerized Adaptive Testing Functions

Item Parameter Drift

Description

This function identifies items that have become unstable, meaning their item parameter values have shifted, within two specified sets of items.

Usage

ipd(
  ip1,
  ip2,
  method = "robust-z",
  anchor_item_ids = NULL,
  alpha = 0.01,
  iqr_type = 7,
  theta = seq(-4, 4, 0.1),
  weights = stats::dnorm(seq(-4, 4, 0.1))
)

Arguments

ip1	An Itempool object for the first calibration.
ip2	An Itempool object for the second calibration.
method	The method for analyzing item parameter drift. "robust-z" Robust-Z method based on the Huynh and Meyer (2010). "d2" D2 method assesses item parameter drift using the method outlined in Wells et al. (2014). It involves comparing the Item Characteristic Curves (ICCs) of item parameters from two different item pools. This is also referred to as WRMSD (Weighted Root Mean Squared Difference). There are no strict thresholds for determining the significance of D2 in identifying item drift. A comprehensive approach considering other measures is recommended. Nevertheless, as a general guideline, for dichotomous items, a D2 value greater than 0.1 may warrant further scrutiny. For polytomous items with two thresholds (or three score categories), a D2 value exceeding 0.15, for those with three thresholds (or four score categories), a D2 value greater than 0.225, for those with four thresholds (or five score categories), a D2 value larger than 0.3, and for items with five thresholds (or six score categories), a D2 value larger than 0.375 may be indicative of item drift and should be investigated further.
anchor_item_ids	A character vector containing the IDs of anchor items. If set to NULL, it is assumed that all items are considered anchor items.
alpha	A numeric value ranging from 0 to 1. Only needed when method = "robust-z". The two-tailed critical value is employed to identify unstable items. For instance, if we calculate the critical value using qnorm(1-alpha/2) (which equals 1.96 when alpha = 0.05), items with absolute robust-z values exceeding this threshold will be marked as unstable.
iqr_type	An integer indicating the choice of quantile algorithm. Refer to the ?quantile function's type argument for more details. For instance, SAS's default quantile algorithm, QNTLDEF=5, corresponds to iqr_type = 2 in R. The default value is iqr_type = 7.
theta	A numeric vector containing the quadrature points. Only needed when method = "d2".
weights	A numeric vector containing the weights assigned to the quadrature points. The length of this vector should match the length of the theta argument. Only needed when method = "d2"

Value

Return a list depending on the method:

robust-z

output$a$cor

Correlation between two sets of a parameters.

output$a$sd_ratio

The ratio of the standard deviation of ip2 to the standard deviation of ip1.

output$a$robust_z

Robust-z statistic values for each item's discrimination parameter.

output$a$unstable

Item IDs that were flagged when the robust-z statistic value for a parameters exceeded the absolute value of the critical value (i.e., qnorm(1-alpha/2)).

output$b$robust_z

Robust-z statistic values for each item's difficulty or threshold parameter. If an item has multiple threshold parameters, robust z statistics will be calculated for each one.

output$b$unstable

Item IDs that were flagged if the robust-z statistic for difficulty/threshold parameters exceeded the absolute value of the critical value (i.e., qnorm(1-alpha/2)).

d2

A numeric vector containing the differences between the ICCs of each item.

Author(s)

Emre Gonulates

References

Huynh, Huynh and Meyer, Patrick (2010) "Use of Robust z in Detecting Unstable Items in Item Response Theory Models," Practical Assessment, Research, and Evaluation: Vol. 15 , Article 2. <doi:10.7275/ycx6-e864>

Examples

##### Robust-z #####
# The example from Huynh and Meyer (2010)
ip1 <- c(itempool(
  a = c(0.729, 0.846, 0.909, 0.818, 0.742, 0.890, 1.741, 0.907, 1.487, 1.228,
        0.672, 1.007, 1.016, 0.776, 0.921, 0.550, 0.624, 0.984, 0.506, 0.594,
        0.687, 0.541, 0.691, 0.843, 0.530, 0.462, 1.007, 0.825, 0.608, 1.177,
        0.900, 0.861, 0.843, 1.404, 0.446, 1.014, 1.632, 0.831, 1.560, 0.798),
  b = c(1.585, 0.635, -0.378, -0.100, -0.195, 0.749, 1.246, 1.016, -0.234,
        0.537, 0.070, 1.985, 1.101, -0.742, 0.463, -0.060, 0.477, 1.084,
        -2.340, 1.068, -0.055, -1.045, 1.859, 0.645, -0.689, -2.583, 1.922,
        0.709, 0.499, 1.973, 0.104, 0.809, 0.640, 0.247, 0.820, 1.837,
        2.129, 1.012, 1.774, 0.095),
  c = c(0.134, 0.304, 0.267, 0.176, 0.215, 0.194, 0.267, 0.159, 0.095,
        0.197, 0.089, 0.272, 0.229, 0.159, 0.162, 0.100, 0.259, 0.167,
        0.000, 0.242, 0.323, 0.000, 0.196, 0.189, 0.000, 0.000, 0.334,
        0.538, 0.125, 0.511, 0.192, 0.353, 0.103, 0.241, 0.245, 0.118,
        0.155, 0.132, 0.215, 0.148),
  model = "3PL"),
  item(a = 0.561, b = c(0.784, -0.113, 1.166), model = "GPCM"),
  item(a = 0.745, b = c(3.687, 2.506, -0.001), model = "GPCM"))

ip2 <- c(itempool(
  a = c(0.650, 0.782, 0.816, 0.787, 0.611, 0.888, 1.192, 0.589, 1.211,
        0.742, 0.526, 0.690, 0.996, 0.816, 0.781, 0.507, 0.378, 0.976,
        0.473, 0.364, 0.585, 0.566, 0.511, 0.718, 0.354, 1.080, 0.840,
        0.865, 0.528, 0.814, 0.555, 0.701, 0.530, 1.220, 0.344, 0.966,
        1.044, 0.358, 1.192, 0.615),
  b = c(0.676, -0.525, -1.749, -1.092, -1.619, -0.406, -0.132, 0.006,
        -1.352, -0.872, -1.242, 0.873, 0.239, -2.038, -0.487, -1.372,
        -1.492, 0.214, -4.537, 0.220, -0.686, -2.394, 0.747, -0.467,
        -3.629, -5.000, 0.927, 0.305, -0.839, 1.270, -1.618, -0.091,
        -1.228, -1.019, -1.453, 1.090, 1.743, -1.436, 1.024, -1.358),
  c = c(0.110, 0.316, 0.161, 0.149, 0.145, 0.200, 0.243, 0.059, 0.081,
        0.075, 0.028, 0.267, 0.242, 0.189, 0.184, 0.121, 0.000, 0.170,
        0.000, 0.151, 0.383, 0.000, 0.195, 0.177, 0.000, 0.000, 0.352,
        0.647, 0.116, 0.501, 0.000, 0.286, 0.000, 0.248, 0.064, 0.150,
        0.126, 0.000, 0.187, 0.007),
  model = "3PL"),
  item(a = 0.486, b = c(-0.539, -1.489, -0.052), model = "GPCM"),
  item(a = 0.737, b = c(2.599, 1.250, -1.209), model = "GPCM"))
ipd(ip1, ip2)

##### D2 #####
ip1 <- generate_ip(n = 20)
ip2 <- ip1
# add a small nuisance to item difficulty parameters
ip2$b <- ip1$b + runif(20, -.5, .5)

theta <- seq(-4, 4, 0.2)
weights <- dnorm(theta)
ipd(ip1, ip2, theta = theta, weights = weights)

# Calculate for only certain items
ipd(ip1, ip2, theta = theta, weights = weights,
  anchor_item_ids = c("Item_2", "Item_6", "Item_9", "Item_13"))

### Polytomous items items
n_item <- 30
models <- sample(c("3PL", "GPCM2"), n_item, TRUE)
new_ip <- generate_ip(model = models, D = 1.702)
old_ip <- data.frame(new_ip)
old_ip$a <- old_ip$a + round(runif(n_item, min = -.5, max = .5), 2)
old_ip$b <- old_ip$b + round(runif(n_item, min = -.75, max = .75), 2)
old_ip$d1 <- old_ip$d1 + round(runif(n_item, min = -.75, max = .75), 2)
old_ip$d2 <- old_ip$d2 + round(runif(n_item, min = -.75, max = .75), 2)
old_ip$d3 <- old_ip$d3 + round(runif(n_item, min = -.75, max = .75), 2)
old_ip <- itempool(old_ip)

ipd(ip1 = old_ip, ip2 = new_ip, theta = theta, weights = weights)

irt documentation built on May 29, 2024, 12:02 p.m.