We investigate double $\Lambda$ hyperfragment formation from the statistical decay of double $\Lambda$ compound nuclei produced in the $\Xi^-$ absorption at rest in the light nuclei $^{12}\mathrm{C}$, $^{14}\mathrm{N}$, and $^{16}\mathrm{O}$. We examine the target and the $\Lambda\Lambda$ bond energy dependence of the double $\Lambda$ hyperfragment formation probabilities, especially of those double hypernuclei observed in experiments. For the $^{12}\mathrm{C}$ ($^{14}\mathrm{N}$) target, the formation probabilities of $^{\,\;\;6}_{\Lambda\Lambda}\mathrm{He}$ and $^{\;10}_{\Lambda\Lambda}\mathrm{Be}$ ($^{\;13}_{\Lambda\Lambda}\mathrm{B}$) are found to be reasonably large as they are observed in the KEK-E373 (KEK-E176) experiment. By comparison, for the $^{16}\mathrm{O}$ target, the formation probability of $^{\;11}_{\Lambda\Lambda}\mathrm{Be}$ is calculated to be small with $\Delta B_{\Lambda\Lambda}$ consistent with the Nagara event. We also evaluate the formation probability of ${}^{\,\;\;5}_{\Lambda\Lambda}\mathrm{H}$ from a $\Xi^-$–${}^{6}\mathrm{He}$ bound state, ${}^{7}_{\Xi}\mathrm{H}$.