CODEWITHSUNDEEP
rstatisticslongitudinal
MervS
MervS

Reputation: 3

Using GAMs to predict change in longitudinal data using covariates scores at a single time-point in R

I have a question on analyzing longitudinal data in R.

To provide a little context, my dataset is in long format, organized for each subject ('ID') across 3 time-points. Importantly, my data is unbalanced (with some complete observations across 3 waves, whilst others may only have data for 1 or 2 waves). A simplistic example is presented below:

ID   timepoint     outcome      predictor  

001     1            244           100       
001     2            305           144
002     1            122           200
002     2            266           120
002     3            308           118
003     2            311           129
003     3            411           126

I'm planning to run a generalized additive model (via the mgcv package) to examine whether change in my continuous DV ('outcome') across the 3 waves can be significantly predicted using scores only from timepoint 1.

So essentially my desired model will be look like the following:

model1 = gam(score ~ s(predictor_scores_at_timepoint1, k=4), data=df, method='ML')

Is there an intuitive way to go about this?

Many Thanks! :)

Upvotes: 0

Views: 143

Answers (1)

Brian Montgomery
Brian Montgomery

Reputation: 2414

I think the easiest is to divide your data into train and test.

train <- df[timepoint == 1, ]
test <- df[timepoint > 1, ]

Then run your gam:

model1 <- gam(outcome ~ s(predictor, k = 4), data = train, method = 'ML')

Then predict:

test$predictions <- predict(model1, newdata = test, type = 'response')

Upvotes: 0

Related Questions