Researchers have been making use of ecological momentary assessment (EMA) and other study designs that sample feelings and behaviors in real time and in naturalistic settings to study temporal dynamics and contextual factors of a wide variety of psychological, physiological, and behavioral processes. As EMA designs become more widespread, questions are arising about the frequency of data sampling, with direct implications for participants' burden and researchers' ability to capture and study dynamic processes. Traditionally, spectral analytic techniques are used for time series data to identify process speed. However, the nature of EMA data, often collected with fewer than 100 measurements per person, sampled at randomly spaced intervals, and replete with planned and unplanned missingness, precludes application of traditional spectral analytic techniques. Building on principles of variance partitioning used in the generalizability theory of measurement and spectral analysis, we illustrate the utility of multilevel variance decompositions for isolating process speed in EMA-type data. Simulation and empirical data from a smoking-cessation study are used to demonstrate the method and to evaluate the process speed of smoking urges and quitting self-efficacy. Results of the multilevel variance decomposition approach can inform process-oriented theory and future EMA study designs.