global_fit.qmd

---
format: html
---

# Study 6: Tests of global fit 

Tests of global fit are often used in published psychometric research...

```{r}
pcasim <- function(dat, iterations, samplesize, cpu = 9) {
  
  require(doParallel)
  registerDoParallel(cores = cpu)
  
  fit <- data.frame()
  fit <- foreach(i = 1:iterations, .combine = rbind) %dopar% {
    data <- dat[sample(1:nrow(dat), samplesize), ]
    
    erm_out <- RM(data)
    ple <- eRm::person.parameter(erm_out)
    item.fit <- eRm::itemfit(ple)
    std.resids <- item.fit$st.res
    
    pca <- pca(std.resids, nfactors = ncol(data), rotate = "oblimin")
    pca$values[1]
  }
  return(fit)
}

test <- pcasim(simdata3, 100, 500, cpu = 1)
# expected values when the data fits, no misfitting items
baseline_pca <- pcasim(simdata[[1]][,-9], 100, 500, cpu = 1)
```

```{r}
hist(test)
hist(baseline_pca)
```


```{r}
lrtsim <- function(dat, iterations, samplesize, cpu = 9) {
  
  require(doParallel)
  registerDoParallel(cores = cpu)
  
  fit <- data.frame()
  fit <- foreach(i = 1:iterations, .combine = rbind) %dopar% {
    data <- dat[sample(1:nrow(dat), samplesize), ]
    
    erm_out <- RM(data)
    lrt <- LRtest(erm_out)
    lrt[["pvalue"]]
    # same p-values with clr_tests()
    #clr <- clr_tests(dat.items = data, model = "RM")
    #clr[,3]
    
  }
  return(fit)
}

test_lrt <- lrtsim(simdata3, 100, 500, cpu = 1)
test_lrt2 <- lrtsim(simdata3, 100, 500, cpu = 1)

# expected values when the data fits, no misfitting items
baseline_lrt <- lrtsim(simdata[[1]][,-9], 100, 500, cpu = 1)
```

```{r}
hist(as.numeric(test_lrt))
hist(test_lrt2)
hist(as.numeric(baseline_lrt))
```