Benchmarking the benchmarks: A mixture-of-experts model of heterogeneous reference points in laptop choice

Consumers rarely evaluate multi-attribute products on absolute levels alone; instead, they judge each option relative to salient reference points — either external benchmarks (e.g., the best, worst, or average option available) or internal anchors based on personal experience (e.g., a previously owned product). Drawing on reference-dependent choice theory and context-effect research, we propose a Mixture-of-Experts (ME) choice model that captures heterogeneous reference-point strategies. In this model, each reference type serves as an “expert” with its own utility function, and a gating mechanism assigns individuals to these experts based on consumer-specific characteristics. We test this approach using an online laptop conjoint experiment including an opt-out (no-choice) option. The ME model yields better fit and predictive accuracy than traditional attribute-only or single-reference models and provides behavioral insights. The best option often acts as an aspirational anchor; internal and average references are frequently used, whereas worst-case comparisons are uncommon. Reference framing further alters attribute weights—for example, consumers show steeper price sensitivity when evaluated against an internal anchor—underscoring how context and personal anchors jointly shape preferences. We outline practical implications, such as highlighting “best-in-class” options for those drawn to top benchmarks, framing upgrades for consumers anchored on prior products, designing balanced configurations for shoppers who favor average options, and using the estimated opt-out constant to diagnose deferral thresholds. Overall, accounting for which reference strategy each user employs—and who tends to use each—improves the explanation and prediction of complex product choices and offers actionable guidance for user interface design, marketing, and personalized recommendation systems. • MoE model captures heterogeneous reference-point strategies in laptop choice. • Best and internal anchors drive choices; worst comparisons are uncommon. • Gate links strategy use to user traits, revealing behavioral segments. • Reference framing shifts price and feature sensitivities and deferral thresholds. • Implications for interface framing and transparent, personalized recommendations.