How to deal with small sample sizes at the group-level for Dyadic Data Analysis in Speed-Dating Contexts?