In this paper, we contribute to the theoretical understanding, analysis, and design of spatial modulation multiple-input-multiple-output (SM-MIMO) systems for transmit diversity without channel state information at the transmitter. The contribution is threefold: 1) The achievable transmit diversity of SM-MIMO is analytically studied by analyzing the impact of various design parameters, notably spatial constellation diagram and shaping filters at the transmitter; 2) the design of SM-MIMO providing transmit diversity and maximum-likelihood (ML) optimum single-stream decoding is investigated; and 3) via Monte Carlo simulations, a comprehensive performance assessment of SM-MIMO against state-of-the-art MIMO (e. g., spatial multiplexing, orthogonal space-time block codes, Golden code, and double space-time transmit diversity) is conducted. It is shown that, for many system setups, a properly designed SM-MIMO outperforms, with lower decoding complexity, state-of-the-art MIMO. In particular, SM-MIMO is particularly useful in the downlink, where many antenna elements (with only few of them active) are available at the transmitter, and few antenna elements are available at the receiver.