Why are observables hermitian operators in the Everett interpretation? Observables correspond to hermitian operators on the quantum state. 
But in the Everett interpretation, the wave function doesn’t collapse since we consider the entire universe as a single quantum state at time $t$, so observation only happens “within that quantum state” (or something like that). 
This confuses me. Given that observations are not treated differently than other events in the Everett interpretation, how does this relate to observations being hermitian operators?
 A: Even in traditional interpretations of quantum mechanics it is easier to define what is an observable than what is an observation: it is a property of a quantum system or subsystem that we in principle have access to through observation (whatever the latter means). 
If we say that this is a decomposition in orthogonal subspaces of the state space, which are distinguished by a real numerical value assigned to it, called the outcome of the measurement, as does the measurement postulate, we don't need to know anymore what we mean by an observation.
In Everett's interpretation an observation is described by unitary evolution just like everything else. A decomposition of the state space in orthogonal subspaces however remains meaningful, just like in traditional interpretations. Let's still call the associate number an outcome. Disregarding subtleties when working in infinite dimensional state spaces, orthogonal decompositions in subspaces indexed by some real numbers are exactly equivalent to hermitian operators (just pick your orthonormal basis $|\lambda\rangle$ with corresponding outcome $\lambda$. The corresponding hermitian operator is $\sum\lambda|\lambda\rangle\langle\lambda|$. This happens to be one of the occasions where I really like the Dirac notation). It could be argued that the former, the decomposition with outcomes, actually is the more fundamental definition of an observable. 
A: While many worlds interpretation provides a complete and thorough description of the state of quantum systems (including observers as quantum system), it doesn't provide a prescription for how the physical state of the system is related to the experience of observers.
In many worlds there is no significance to any operators on the Hilbert space unless those operators happen to be part of the Hamiltonian (which is Hermitian) which is used to determine the time evolution of the universal wavefunction.
Otherwise there is no special significance to Hermitian operators. One might say that many worlds doesn't give a prescription for how observations happen so there are no such thing as observations in the many worlds interpretation and thus there is no need for observables.
To answer your question directly

Given that observations are not treated differently than other events in the Everett interpretation, how does this relate to observations being hermitian operators?

Since observations are treated no differently than other events (unitary evolution through the Schrodinger equation for all events) there are no observation so there is no relationship between observations and hermitian operators.
Because MWI doesn't give a prescription relating our experiences to the physical states I see it as being a fundamentally incomplete and scientifically insufficient interpretation of quantum mechanics. I would think that anyone who says otherwise is implicitly or explicitly adding additional postulates to what is usually understood as the many worlds interpretation, that is pure wave mechanics.
A: A measurement is an interaction with a system that produces information about that system that can be copied, like an entry in a lab book, or a spreadsheet or a database or whatever. In quantum mechanics without collapse, this constraint requires that the information being copied is represented by a sum of an orthonormal set of projectors, as explained by Zurek:
https://arxiv.org/abs/1212.3245
It seems reasonable to regard the eigenvalues as the results of the measurement and to require that the eigenvalues are real since complex numbers are used to describe interference.
This view fits in with David Deutsch's explanation of universes in the Everett interpretation as channels in which information flows:
https://arxiv.org/abs/quant-ph/0104033
A: Real observables are hermitic operators always. If they were not, it would be eigenstates from this operator that would have complex values. No matters the interpretation
