Modeling Cyclical Desire: Markov Chains vs. Renewal Theory

1. Modeling Recurrent Encounters: State vs. Event

Let's consider the hotwife's series of intimate encounters with her bull. We want to model the timing of these encounters. There are two primary ways to conceptualize this process:

• A State-Based View (Markovian): We could define a set of discrete "desire states," such as Low Desire, Building Desire, and Peak Desire/Ready for Encounter. The process moves between these states over time. A Markov model would assume the probability of moving to the next state depends only on the current state. A key consequence, particularly in continuous time, is that the duration spent in any given state is memoryless—it follows an exponential distribution. The hazard rate of transitioning out of a state is constant.

• An Event-Based View (Renewal): We can focus directly on the encounters themselves as the "events" of interest. Instead of modeling the states between events, we model the time elapsed between consecutive events. This is the domain of Renewal Theory.

The fundamental choice between these approaches hinges on the assumptions we are willing to make about the process's memory.

2. The Essence of Renewal Theory

In the previous lesson, we established that a Markov process is memoryless at every point in time. The future is conditionally independent of the past, given the present state. Renewal Theory offers a different perspective on memory.

A renewal process is a sequence of events where the time intervals between consecutive events are independent and identically distributed (IID) random variables.

The memory property of a renewal process can be summarized as:

The process is memoryless only at the moment an event occurs.

After an encounter happens (a "renewal"), the stochastic process governing the time until the next encounter begins anew, independent of the history of all previous encounters. However, within the cycle—the time between two encounters—the process can have a rich structure. The probability of an event happening can, and usually does, depend on the time that has elapsed since the last one.

This is the critical distinction:

• Markov Chain: Memoryless based on the current state.
• Renewal Process: Memoryless based on the last event time.

3. Choosing the Right Model for Desire's Cycle

The choice of model should be driven by the phenomenon we are trying to capture. A paper on modeling recurrent events in epidemiology provides an excellent framework for this decision, which we can adapt directly to our intimacy scenario.

Let's use this framework to compare the models for the hotwife's encounters.

Case 1: The Markov Assumption (Constant Hazard)

As discussed in the LINK resource, a model like the Andersen-Gill (AG) model is suitable when "the risk of recurrent events remained constant regardless of the number of previous events." In our context, this would be analogous to using a simple continuous-time Markov Chain or a Poisson process (which we'll discuss next lesson).

• Underlying Assumption: The likelihood of an encounter happening in the next small time interval is constant, regardless of how long it has been since the last encounter. This implies the time between encounters follows an exponential distribution.
• Plausibility: Is this realistic for human desire? It suggests that immediately after an encounter, the probability of another one is just as high as it is a week later. This seems unlikely. Most biological and psychological cycles involve periods of latency, buildup, and saturation. A constant-hazard model fails to capture this internal dynamic.

Case 2: The Renewal Theory Assumption (Variable Hazard)

The PWP-GT model, which assumes a renewal process, is recommended when the risk of recurrence changes. This is the key advantage of Renewal Theory for our scenario.

• Underlying Assumption: The time between encounters is an IID random variable, but its distribution can be arbitrary. It does not have to be exponential.
• Plausibility: This is far more flexible and realistic. We can select a distribution for the inter-encounter time that reflects a plausible story about the renewal of desire.
  • Weibull Distribution: If the shape parameter $k>1$, the hazard rate increases with time. This models a situation where the longer the time since the last encounter, the more "pressure" or desire builds, making the next encounter progressively more likely.
  • Gamma or Log-Normal Distribution: These distributions can model a hazard rate that first increases and then decreases. This could represent an initial "refractory" or low-desire period, followed by a buildup to peak desire, after which, if an encounter doesn't occur, the intensity might wane due to other life factors.

Justification: For modeling the cyclical renewal of desire, Renewal Theory is the more suitable framework. Its strength lies in decoupling the memorylessness between cycles from the dynamics within a cycle. It allows us to use a rich family of probability distributions to model the time-to-next-encounter, capturing plausible psychological dynamics like desire buildup, which a simple Markov model with its constant hazard assumption cannot.

4. Connecting Markov Chains and Renewal Theory

While we've contrasted the two, it's important to understand they are not entirely separate. In fact, Renewal Theory is a generalization.

A process where events arrive according to a Poisson process is a special case of a renewal process where the interarrival times are exponentially distributed. This is the "most random" or "most memoryless" renewal process, and it connects directly to continuous-time Markov Chains.

Furthermore, we can use renewal concepts to analyze Markov Chains. Consider a complex Markov model of desire with many states. The sequence of times at which the process returns to a specific state (e.g., Peak Desire) forms a renewal process.

This reveals a powerful idea: even if you model the moment-to-moment dynamics with a Markov Chain, the long-term recurrence of key events (like encounters) can be analyzed using the powerful tools of Renewal Theory. A more advanced model, a Semi-Markov Process, formalizes this by allowing the time spent in each state to follow an arbitrary distribution, effectively creating a bridge between the two theories.

Conclusion

In this lesson, we compared Markov Chains and Renewal Theory as frameworks for modeling the recurrent cycle of intimate encounters. Your background in statistics makes you well-equipped to appreciate the nuances in their respective assumptions about memory.

Key Takeaways:

• Markov Chains assume a process is memoryless at all times, conditional on the present state. This implies constant hazard rates and exponentially distributed waiting times in states, which can be an overly strong assumption for many real-world cycles.
• Renewal Theory assumes a process is memoryless only at event times. It focuses on the IID nature of the time between events, allowing this time to follow any probability distribution.
• For modeling the renewal of desire, Renewal Theory is generally more suitable because it provides the flexibility to model non-constant hazard rates, capturing realistic dynamics like a buildup of desire over time.
• The choice of model is a pragmatic one based on plausibility. The more complex, non-exponential patterns of human behavior are often better captured by the more general renewal framework.

In our next lesson, we will put this theory into practice. We will focus on the simplest renewal model—the Poisson process—and learn how to test the goodness-of-fit of a Poisson process for the arrival of 'peak photogenic moments' during a shoot, and analyze the implications if arrivals are non-Poisson. This will give us a concrete statistical tool to determine if the simple "constant hazard" assumption is justified by data.