This paper investigates state dependence in self-curated experiences, those where individuals sequentially select and consume a series of unique options in real time. Leveraging large-scale data from Spotify listening sessions, complemented by experimental evidence, this paper finds that consumers exhibit positive state dependence in these. Consumers are more likely to skip a song, exit a session, and report lower enjoyment the more dissimilar a song is from previously consumed songs. Moreover, there is a decay in this effect, such that the distance to the most recently played song has a stronger effect than the distance to songs played 2-back and 3-back. Finally, we find that consumer effort (actively rejecting options) results in less consumed variety overall, indicating that active curation narrows rather than expands experiential variety.

Starting from over 2,000 papers identified through systematic keyword search, I used a combination of an AI assisted screening and two independent human raters to arrive at a final sample of 132 papers with extractable effect estimates. The analysis surfaces two robust patterns. First, the temporal context matters: state dependence tends to be negative within session — consumers seek variety in a short continuous consumption context — but positive between sessions, where habit and loyalty dominate. Second, the consumption type matters: experiential products such as music and video games exhibit more negative state dependence, while CPG categories and tobacco exhibit more positive state dependence. Together, these patterns suggest that the classical “positive state dependence” finding from CPG brand-choice research generalizes poorly to online streaming settings.

This study examines the impact of large-scale subsidy campaigns on consumer loyalty in a two-sided multi-homing context, where both business owners and consumers simultaneously participate on multiple platforms. We ask whether temporary price subsidies in one category can generate lasting changes in consumer purchase behavior, and whether these habits spill over into other categories even in the absence of subsequent promotions. We collaborate with a third party company with order-level data from multiple major instant retailing platforms, and build a dynamic competition model that incorporates consumers’ purchase history as a state-dependent factor. By modeling platform–category state dependence, we identify platform-level stickiness that extends beyond traditional brand-level state dependence, while also capturing potential cross-category spillover effects. For identification, we exploit an exogenous shock to prices arising from a subsidy war, leveraging its staggered rollout and differential intensity across platforms and categories. The results help managers assess the overall return on subsidy campaigns.

This project develops a theoretical foundation for the search-and-consumption process underlying real-time sequential consumption. I extend the classic secretary problem along two dimensions that match the streaming environment: recursive stopping, where the agent re-enters the search process after each consumption episode rather than stopping once, and non-recall, where previously skipped options cannot be revisited. The model captures consumption on platforms like music and short-video services, where users sequentially evaluate options, consume the ones they select, and continue searching afterward — never returning to what they skipped. Because the agent re-enters search after each episode, the outcome of one consumption decision becomes part of the state governing the next: the reservation threshold is no longer constant but evolves with what was just consumed. This makes state dependence endogenous to the search process — positive when prior consumption raises the continuation value of similar options (loyalty, inertia), negative when it depresses the marginal value of the next (satiation, variety-seeking). The framework thus shows how consumers allocate attention across a stream of experiential options, generates testable predictions about how stopping behavior shifts with content-quality dispersion, consumption cost, and expectations about future options, and offers a behavioral micro-foundation for the state dependence that discrete-choice models estimate but do not explain.