ads2-ica-open-analysis

name

ads2-ica-open-analysis

description

Use this skill for ADS2 ICA or open-ended analysis reports, especially substance_use.csv, Global Burden of Disease data, filtering by measure/location/sex/age/cause/year, trend plots, age/sex comparisons, asking an original question, reproducible RMarkdown, and interpreting observational data. Trigger on ICA, group exercise, substance_use, open question, prevalence, deaths, opioid, alcohol, trend over time, or "ask your own question".

ADS2 ICA Open Analysis

ICA marking emphasizes clarity, complete code, interpretation, reproducibility, and sensible scope. A simple question thoroughly answered can score better than an ambitious weak model.

ICA Workflow

1. State the question.
2. Import and check data.
3. Filter relevant rows.
4. Summarize values with grouping.
5. Plot before modeling.
6. Choose a course-covered method.
7. Interpret in context.
8. State limitations and a next step.

Substance Use Dataset Columns

Common columns:

• measure: Deaths or Prevalence
• location: world region
• sex: Male or Female
• age: 5-year age group
• cause: Alcohol use disorders or Opioid use disorders
• metric: usually Percent
• year
• val: best estimate
• upper, lower: uncertainty interval

Import And Check

library(tidyverse)
health_df <- read.csv("substance_use.csv")
str(health_df)
head(health_df)
colSums(is.na(health_df))
sum(duplicated(health_df))

Do not use personal absolute paths in final submission.

Direct Question Example

"In 2019, what region has the highest alcohol-related death rate among men aged 40-44?"

q1 <- health_df |>
  filter(measure == "Deaths",
         cause == "Alcohol use disorders",
         sex == "Male",
         age == "40 to 44",
         year == 2019)

q1[which.max(q1$val), c("location", "val", "lower", "upper")]

Trend Plot

trend <- health_df |>
  filter(measure == "Prevalence",
         cause == "Opioid use disorders",
         location == "North America") |>
  group_by(year, age) |>
  summarise(mean_val = mean(val), .groups = "drop")

ggplot(trend, aes(x = year, y = mean_val, color = age)) +
  geom_line() +
  labs(x = "Year", y = "Prevalence (%)")

Regression-By-Group Trend

models <- trend |>
  group_by(age) |>
  do(model = lm(mean_val ~ year, data = .))

slopes <- sapply(models$model, function(m) coef(summary(m))[2, 1])
data.frame(age = models$age, slope = slopes)

Comparing Periods

early <- trend |> filter(year >= 1992, year <= 1997)
late <- trend |> filter(year >= 1998, year <= 2003)

get_slopes <- function(dat) {
  models <- dat |>
    group_by(age) |>
    do(model = lm(mean_val ~ year, data = .))
  sapply(models$model, function(m) coef(summary(m))[2, 1])
}

slopes_early <- get_slopes(early)
slopes_late <- get_slopes(late)
wilcox.test(slopes_late, slopes_early, paired = TRUE, alternative = "greater")

Good Open Question Criteria

Choose a question that:

• uses only available columns or clearly cited additional data;
• can be answered by plots, summaries, hypothesis tests, regression, or simulation learned in the course;
• has enough data after filtering;
• has a clear response variable and comparison/trend;
• avoids causal claims from observational data.

ICA Interpretation Template

The plot shows [trend/pattern] in [group/location]. The model/test provides [evidence/insufficient evidence] that [specific claim]. Because these are observational prevalence/death estimates, this should be interpreted as an association/trend rather than proof of causation. A useful next step would be [specific follow-up].