Cluster analysis of the NDSR Competencies survey (https://osf.io/zndwq/)

ndsr_skill_cluster.r

library(readr)
library(dplyr)
library(tidyr)
library(factoextra)

# Read data, drop last 3 columns that don't ask about skills
osf_url = "https://files.osf.io/v1/resources/zndwq/providers/osfstorage/570be0eab83f6901d62b19d9"
responses <- read_csv(osf_url)[,1:30]

# Reshape raw data into number of responses per importance level
responses %>% 
  # Fix typo
  rename(NeedsAssessGapAnalyses = NeedsAssessGayAnalyses) %>%
  gather(skill) %>%
  group_by(skill, value) %>%
  summarize(n = n()) %>%
  ungroup() %>%
  spread(value, n, fill = 0) %>%
  # Move back to data.frame since tibbles can't have row labels
  as.data.frame()->
  skills

# Add skill names as row labels and drop skill column
rownames(skills) <- skills$skill 
skills <- skills[,-1]

# Run cluster analysis with repeatable seed
set.seed(42)
skill_clusters <- eclust(skills, "kmeans", k = 3, nstart = 30, graph = F)

# Visualize k-means clusters
fviz_cluster(skill_clusters, ellipse.type = "norm", repel = T) +
  labs(x = "55% of variance in necessity", y = "25% of variance in necessity",
       title = "NDSR Project Skills Grouped by Variance of Importance") +
  # Add meaning to cluster labels, wish there was a convenience function to do this
  scale_shape_discrete(name="Skill Need",
                       breaks=c(1, 2, 3),
                       labels=c("Required Skills", "Niche Skills", "Nice-to-Have Skills")) +
  scale_fill_discrete(name="Skill Need",
                      breaks=c(1, 2, 3),
                      labels=c("Required Skills", "Niche Skills", "Nice-to-Have Skills")) +
  scale_color_discrete(name="Skill Need",
                       breaks=c(1, 2, 3),
                       labels=c("Required Skills", "Niche Skills", "Nice-to-Have Skills")) + 
  ggsave('ndsr_skill_cluster.png')