We explore a variety of statistics of clusters selected with cosmic shear measurement by utilizing both analytic models and large numerical simulations. We first develop a halo model to predict the abundance and the clustering of weak lensing selected clusters. Observational effects such as galaxy shape noise are included in our model. We then generate realistic mock weak lensing catalogues to test the accuracy of our analytic model. To this end, we perform full-sky ray-tracing simulations that allow us to have multiple realizations of a large continuous area. We model the masked regions on the sky using the actual positions of bright stars, and generate 200 mock weak lensing catalogues with sky coverage of ∼1000�deg². We show that our theoretical model agrees well with the ensemble average of statistics and their covariances calculated directly from the mock catalogues. With a typical selection threshold, ignoring shape noise correction causes overestimation of the clustering of weak lensing selected clusters with a level of about 10 per cent, and shape noise correction boosts the cluster abundance by a factor of a few. We calculate the cross-covariances using the halo model with accounting for the effective reduction of the survey area due to masks. The covariance of the cosmic shear auto power spectrum is affected by the mode-coupling effect that originates from sky masking. Our model and the results can be readily used for cosmological analysis with ongoing and future weak lensing surveys.