simulating what if scenarios using Bayesian model - rstanarm

Question

I am fitting a model (very much simplified for reproducible) like so:

library(datasets) library(rstanarm)

set.seed(42)
rm(list = ls(all = TRUE))

prediction_data1 <- data.frame(
        Petal.Length = 1.4
    )

prediction_data2 <- data.frame(
        Petal.Length = 1.4
    )

model <- stan_glm(
        Petal.Width ~ Petal.Length
        , data = iris
        , chains = 3
        , iter = 1000
        , warmup = 100
)

new_predictions1 <- as.data.frame(posterior_predict(model, newdata = prediction_data1))
new_predictions2 <- as.data.frame(posterior_predict(model, newdata = prediction_data2))

colnames(new_predictions1) <- c('Petal.Width')
colnames(new_predictions2) <- c('Petal.Width')

median(new_predictions1$Petal.Width)
median(new_predictions2$Petal.Width)

I guess I should not expect the same median for both equal 'simulation' datasets (e.g. 0.2216209 and 0.2177802)? However is the above skeleton code a correct approach to simulate different scenarios ( e.g. Petal.Length = 1.4 versus Petal.Length = 2.4)?