This study reports experiments with the newly-released CL-Aff HappyDB dataset, which looks beyond positive emotion in modeling descriptions of happy moments collected through writing prompts. The widespread adoption of social media has improved researchers' access to unsolicited expressions and behaviors. However, most of the approaches to analyzing these expressions involve a keyword search and focuses on predicting sentiment or emotional content rather than understanding a deeper psychological state, such as happiness. The CL-Aff HappyDB dataset is the first effort to distinguish the personal agency and social interaction in writings about happiness, which do not yet have an exact equivalent concept in existing text-based approaches. We report that state of the art approaches for emotion detection have different topical characteristics, and do not generalize well to detect happiness in the CL-Aff HappyDB dataset. Language models trained on the dataset, on the other hand, generalize to social media writing and are a valid approach for downstream tasks, such as predicting life satisfaction from social media posts.