Summary Contemporary cohort studies and randomized clinical trials are regularly linked to biorepositories and electronic health records systems. These secondary resources often contain exposure and/or outcome data that are crucial for addressing novel study questions. However, exposure/outcome ascertainment costs are often prohibitive. For example, assaying biospecimen from biobanks to measure blood markers or manually reviewing health records to accurately ascertain medical history information are both costly and restrict sample size. A solution to high ascertainment costs is the two-phase study that uses available participant information to identify those who are most informative for addressing study questions. Restricted study resources are then concentrated on the sub-cohort of informative participants. Outcome dependent and outcome related sampling designs are examples of two-phase studies. They are highly efficient compared to standard random sampling because they use outcome and/or auxiliary variable data to identify the informative participants and then enrich the observed sample with them. However, analyses must correct for the non-representative sample. In this competing renewal, we propose highly efficient outcome dependent and outcome related sampling designs as well as ascertainment correcting analysis procedures for ordinal and longitudinal data. This is a natural extension of the research conducted during the prior funding cycles which focused on longitudinal binary and normally distributed response data. In the current proposal our focus is on generalized ordinal (from a few ordered categories to non-normal, continuous) and longitudinal ordinal responses, on novel semiparametric models, and on robust variations of likelihood-based estimation strategies. Aim 1 regards outcome dependent sampling and outcome related sampling designs and analysis procedures for scalar generalized ordinal response data; Aim 2 regards outcome dependent sampling and outcome related sampling designs and analysis procedures for ordinal, longitudinal data; and Aim 3 extends a new class of semi-parametric generalized linear models (SPGLM) to correlated multi-outcome dependent sampling designs, to longitudinal data settings, and then proposes outcome dependent sampling designs for longitudinal data.