Lucy D’Agostino McGowan

@lucystats.bsky.social

Biostatistician • Associate Prof @ Wake Forest University • former postdoc @ Hopkins Biostat • PhD @ Vandy Biostat • 🎙 Casual Inference • lucymcgowan.com

The workshop attendees and me in a dress that matches my slides!

MMR and V separately, as is common in this age group, is still recommended and available

Trying to run tiny gpt-oss on Mac and failing

roast your colleagues

Jeff Leek’s idea of a wild night is running a bootstrapped simulation — twice.

Git doesn’t give Jenny merge conflicts; it opens an issue with a minimal reprex and apologizes.

Mine’s rubrics are more detailed than most grant proposals — by the time you finish reading them, the semester’s over.

Roger believes in reproducibility — which is great, because his jokes in lecture reproducibly bomb every time

Bad depiction of categorical age having constant effects within year

Plot that shows points with imputed values and the slopes from regression with out without observed values

Plot showing points missing (and not) and the same slope in both even though missing is not at random (with respect to X)

``` r
n <- 10000
effect <- 2
x <- rnorm(n)
y <- effect * x + rnorm(n)
x_obs <- ifelse(x > 0, NA, x)

## truth (fully observed)
lm(y ~ x)
#>
#> Call:
#> lm(formula = y ~ x)
#>
#> Coefficients:
#> (Intercept) x
#> 0.002869 2.012952

## listwise deletion
lm(y ~ x_obs)
#>
#> Call:
#> lm(formula = y ~ x_obs)
#>
#> Coefficients:
#> (Intercept) x_obs
#> -0.003714 2.026903

## mean imputation
x_imp <- ifelse(is.na(x_obs), mean(x_obs, na.rm = TRUE), x_obs)
lm(y ~ x_imp)
#>
#> Call:
#> lm(formula = y ~ x_imp)
#>
#> Coefficients:
#> (Intercept) x_imp
#> 1.639 2.027
```

<sup>Created on 2025-06-25 with [reprex v2.1.1](https://reprex.tidyverse.org)</sup>

$sim <- function() { n <- 100 x1 <- rnorm(n) x2 <- rnorm(n) y <- x1 + x2 + rnorm(n) x1_obs <- ifelse(x1 > 0, NA, x1) x2_obs <- ifelse(x2 > 0, NA, x2) d <- data.frame(y, x1_obs, x2_obs) cc <- broom::tidy(lm(y ~ x1_obs + x2_obs, data = d)) m <- mice::mice(d, print = F, method = "norm") imp <- broom::tidy(mice::pool(with(m, lm(y ~ x1_obs + x2_obs))), conf.int = T) data.frame( cc = cc[2, "std.error"], imp = imp[2, "std.error"] ) } o <- purrr::map_df(1:100, ~sim()) set.seed(1) mean(o$std.err) #> [1] 0.3642946 mean(o$imp) #> [1] 0.3741719$