This paper constructs a theoretical model delineating the connection between day-to-day travel time reliability (TTR) and vehicle density distribution, applying it to freeway stretches in Minneapolis-St. Paul and San Diego for testing. Initially, we establish the presence of a counter-clockwise hysteresis loop phenomenon in the correlation between average vehicle density and experienced travel time standard deviation. We also observe that vehicle density heterogeneity exerts a more profound effect on TTR during moderate congestion. We find that a time-homogeneous Poisson process accurately characterizes the vehicle arrival process on segments of freeway stretches, both with and without on-ramps, during the morning peak period. The travel times for these segments align well with an exponential distribution. Using the M/M/1 queueing model, we derive a model encapsulating the variance of experienced travel time, average vehicle density, and density heterogeneity for freeway stretches consisting of multiple segments. We use the `birth–death process' to model the relationship between average vehicle density and density heterogeneity. By integrating these two models, we derive a novel model that accurately approximates freeway stretch TTR based on average vehicle density and traffic flow. We validate our models by applying them to TTR changes across different time periods on the two selected freeway stretches and by selecting eight additional freeway stretches from San Diego for further testing. The results from these experiments affirm the robustness and generality of our model across both temporal and spatial contexts.