Unlike economic assimilation -- the process of immigrants' wages converging to those of natives -- cultural assimilation -- the process whereby people adopt specific beliefs, languages, or modes of daily life to conform to those around them -- has not been studied in depth using economic tools. This thesis examines cultural assimilation and cultural change as processes resulting from human capital investments. Unlike past studies, where assimilation is viewed as a passive process or as serving to signal identity in a static world, a human capital approach to assimilation emphasizes the productive aspects of culture and nests assimilation into a framework of supply and demand, with intertemporal constraints and forward-looking actors. I focus on three aspects of assimilation that have not been fully examined: the theory of cultural change, empirical evidence for different qualitative types of cultural change, and the proper measurement of cultural human capital.,In the first chapter, I develop a learning-by-doing theory of cultural assimilation, in which an individual's cultural identity is determined by past investments in culture-specific social capital. The model incorporates several empirically relevant aspects of assimilation that have been difficult to explain in past models. The mechanism in my model yields two possible qualitative outcomes of the assimilation process, depending on the intertemporal complementarities of investment. The first, in which individuals become more homogeneous over time, has been used to explain immigrants' wage assimilation. The second, in which individuals become more heterogeneous as some assimilate and some do not, has vastly different implications for immigration and integration policies. As a purely theoretical exercise, the chapter offers a new model of group formation: rather than emerging from individuals forming and erasing costly bonds in a multi-period game, social networks can form endogenously via investment in productive capital.,In the second chapter, I use detailed datasets from three immigrant destination countries to evaluate whether assimilation patterns adhere to the predictions of the first or second type of process from Chapter 1. The data provide a variety of evidence from linguistic, identity, and religiosity measures that cultural assimilation (in contrast to economic assimilation) most closely resembles the second type of process. The regressions typically used on cultural survey data are ordered probit or logistic regressions, or studies will collapse ordinal variables into binary data. A major point of this chapter is that such regressions may be statistically inappropriate, and at the very least can miss important information about the second moment of the distribution -- for example, the fact that there is a long tail of individuals who assimilate very slowly, and who are possibly the most important targets of policy interventions.,In Chapter 3, I construct new measures of linguistic human capital. Using the same data from Chapter 2, I show that different measures of linguistic skills (such as speaking, writing, and reading) measure different dimensions of language fluency. Thus, a proper measure of linguistic skill must incorporate multiple dimensions of evidence. I estimate a continuous latent trait for linguistic skill using a set of 11 survey questions. The methodology comes from the psychometrics literature, designed specifically for analyzing ordinal test data of the sort used to measure language skills. The measures control for systematic differences in the ways different groups answer questions (Differential Item Functioning, or DIF) across groups. I use different latent factor constructs along with standard measures of past studies (binary variables or simple averages of reading/speaking/writing/understanding) to re-estimate the determinants of language skills and the wage premium for learning a language. I show that the results are not sensitive to the exact form of the latent factor, but they are quite different from the results using only binary or discrete variables. In particular, past studies have mis-estimated the wage premium for language skill; I find that there is a premium for speaking, and an additional premium (of almost the same size) for reading/writing in addition to speaking.