Bishop et al. (1975) [39] Review of measures of association for categorical data.
Amemiya (1981) [20] In a review of qualitative response models for economists, some remarks on pseudo-R2 measures, especially for binary models. Here the emphasis is more on model comparison than model summary, so adjustment for degrees of freedom is prioritised over, say, scaling to 0-1.
Agresti (1990) [1] A nice brief summary of R2-measures in a textbook on categorical data analysis.
Windmeijer (1995) [387] Compares all the main pseudo-R2 measures for binary regression models. The main criterion in a simulation is closeness to the R2 for an underlying linear regression (see subsection 8.1.2). A comment by Veall and Zimmermann.
Veall and Zimmermann (1996) [376] A review of essentially all of the pseudo-R2 measures proposed in the literature for `limited dependent variable models'. Recommends that if a latent variable interpretation is sensible, it provides the best criterion and also comparability between different types of models in some situations (e.g. linear and tobit for similar data). Also considers measures of association between predicted and observed categorical outcomes.
Long (1997) [247]
A textbook on regression modelling. Discusses deviance, AIC, BIC,
residuals etc. Also a nice review of various pseudo-$R^{2}$
measures.