Technological change in plant research rarely shifts the entire yield distribution upwards as assumed in the agricultural economics literature. Rather, technologies have been targeted at a specific subpopulation of the yield distribution--for example, drought resistant seeds or so-called racehorse seeds--therefore, it is unlikely technological advancements are equal across subpopulations. In this manuscript we introduce a mixture model of crop yields with an embedded trend function in the component means, which allows different rates of technological change in each mixture or subpopulation. By doing so, we can test some interesting hypotheses that have been previously untestable. While previous literature assumes an equivalent rate of technological change across subpopulations we reject the null in 84.0%, 82.3%, and 64.0% of the counties for corn, soybean, and wheat respectively. Conversely, with respect to stable subpopulations through time (i.e. climate change) we reject in only 12.0%, 5.4%, and 4.6% of the counties for corn, soybean, and wheat respectively. These results have implications for modelling yields, directing funds regarding plant science research, and explaining the prevalence of heteroscedasticity in yield data.