CODEWITHSUNDEEP
rstatisticstime-seriesregression
metaltoaster
metaltoaster

Reputation: 378

Interrupted time series analysis in R; creating textual confirmation of significant results

As part of a Shiny app I am developing, I am creating an "events analysis" tool, of which the basis of is using interrupted time series regression. Here is a reproducible example of interrupted time series analysis:-


library(ggplot2)
library(dplyr)

int = 22
set.seed(42)
df <- data.frame(
  count = as.integer(rpois(132, 9) + rnorm(132, 1, 1)),
  time = 1:132,  
  at_risk = rep(
    c(4305, 4251, 4478, 4535, 4758, 4843, 4893, 4673, 4522, 4454, 4351),
    each  = 12
  )
)


View(df)


df$month <- factor(month.name, levels = month.name)
df$intv <- ifelse(df$time >= int, 1, 0)
df$intv_trend <- c(rep(0, (int - 1)),
                   1:(length(unique(df$time)) - (int - 1)))
df <- df %>%
  mutate(lag_count = dplyr::lag(count))


fit <- glm(count ~ month + time + intv + intv_trend +
             log(lag_count) + offset(log(at_risk)),
           family = "poisson", data = df)


summary(fit)
df$fit <- exp(c(NA, predict(fit)))

# Get predictions on the same scale as the data
df$fit2 = c(NA, predict(fit, type="response"))

# Add a grouping variable manually
df$group = rep(c("Control","Intervention"), c(72, 132 - 72))

ggplot(df, aes(x=time, y = count)) + geom_line()

ggplot(df, aes(x = time, y = fit2)) +
  geom_line() +
  geom_smooth(method="lm", se=F, aes(colour=group)) +
  theme_bw() +
  labs(colour="")


ggplot(df, aes(x = time, y = fit2)) +
  geom_line() +
  geom_smooth(method="loess", se=T, aes(colour=group)) +
  theme_bw() +
  labs(colour="")


ggplot(df, aes(x = time, y = fit2)) +
  geom_point() +
  geom_smooth(method="loess", se=F, aes(colour=group)) +
  theme_bw() +
  labs(colour="")

What I wanted to do was to return a statement telling the user whether a particular predictor variable was statistically significant or not; in this case, the predictor variable in question is "intv". In this model, intv was not significant, so I would like a command that would return a statement, something like the following:-

#if result is significant:-

"The intervention/event had a significant impact on the dependent variable"

#or if it was non-significant:-

"The intervention/event had a non-significant impact on the dependent variable"

Or to that effect at least. Any help is appreciated!

Upvotes: 0

Views: 393

Answers (1)

schoulten
schoulten

Reputation: 68

You can try this:

# Statements

significant <- "The intervention/event had a significant impact on the dependent variable"
non_significant <- "The intervention/event had a non-significant impact on the dependent variable"


# Create a data frame with the model's coefficients and p-values

coef_p <- data.frame(p_value = summary(fit)[["coefficients"]][,"Pr(>|z|)"]) %>%
  tibble::rownames_to_column("coef") %>%
  dplyr::mutate(statement = dplyr::if_else(p_value < 0.05, significant, non_significant, "NA"))

And if I understand correctly, the user must indicate an input (predictor), then you can create an object to store that input and then filter the result (p-value) and print corresponding statement, like this:

predictor = "intv"
result = filter(coef_p, coef == predictor) %>% pull(statement) %>% print(.)

Upvotes: 1

Related Questions