# frequentist influential diagnostics ```r Influence Diagnostics ------------------- rstudent dffits cook.d cov.r QE.del hat weight infl Omitting 1 -1.464 -0.555 0.308 1.144 7.261 0.126 12.558 Omitting 2 2.706 1.861 3.462 1.473 2.081 0.321 32.103 * # wave 2 is influential case Omitting 3 -0.693 -0.656 0.430 1.895 8.923 0.472 47.235 Omitting 4 -1.583 -0.470 0.221 1.088 6.897 0.081 8.104 ``` ![[_temp 25.png]] ## leave-one-out analysis Excluding wave 2 reduces the heterogeneity substantially to just I^2 = 4%. ![[_temp 24.png]] Wave 2 is an influential case. ![[Rplot.png]] # bayesian leave-one-out cross validation many metrics to interpret... but they all consistently agree that wave/row 2 is weird ```r elpd_loo mcse_elpd_loo p_loo looic influence_pareto_k [1,] 1.9307957 0.004916368 0.2508332 -3.8615915 0.2396505 [2,] -0.1668056 0.048454363 1.7375252 0.3336111 0.6848589 # influence [3,] 3.1613302 0.015037437 0.3051857 -6.3226604 0.5688146 [4,] 1.5579564 0.003858606 0.1925829 -3.1159128 0.1701750 ```