For a Binomial Logit Regression, should I run Categorical IV's in a Simple or Multiple Logit Model?

Hi All,

I'm a professor working on a research project that explores the relationship between politicians' facial expressions in social media images and the performance of those posts. I never had much stats training, so looking for help here on whether I should run the IV's in a binary logit regression individually (as a simple logit regression) or together (as a multiple logit regression). I know this has been asked in various threads, but I can't find on answer on the best route when all IV's are categorical and belong to same to the same category.

The DV here is a binary outcome variable - whether a social media post 'overperformed' or not.

The IV's are 8 categories of labelled emotions from the politicians' faces: Happy, Sad, Angry, Unclassified, etc. Each image is only labelled with one emotion classification.

The question is: would it be legit to run each IV as a dummy (e.g., Happy / Non-Happy) in a simple logit regression for each emotion? I assume there's no collinearity between the IVs, because each image is labelled with one and only one emotion.

Or, would it be more accurate to run a multiple logit regression with all IVs at once, or perhaps a stepwise approach to removing all IVs that are non-significant? I suppose here it'd make the most sense to use Unclassified images as the reference category, meaning that coders disagreed that the image contains a clear emotion.

I'm doing this for multiple politicians across two social media platforms, so I'm also thinking about how to streamline the reporting as well. Currently I'm thinking to run simple binomial logits for each IV, and report the results for each emotion as a row, with politicians and the platforms as columns.

Any advice on other approaches would be welcome. This is entirely an exploratory study, so I'm not trying to fit models that predict post performance, necessarily. I'm merely looking for which emotions are most associated with overperforming posts.