Why Driver Weights Differ: A Cost-Based Microfoundation for Segmented Technology Acceptance¶

Stefane Fermigier (Abilian) · sf@abilian.com

Draft working paper v0.3, 2026-09-02. Theory: every proposition carries a proof (Appendix), a numeric cross-check, and a Monte-Carlo robustness report (Section 5). This is the theoretical companion of the acceptance study in stratified-adoption-wp.md, whose H9 states that driver weights differ across capability segments.

Download the PDF

Abstract¶

Acceptance research increasingly compares the driver weights of technology adoption across groups. The companion study's central hypothesis is of this form: effort and facilitating conditions bind hardest for low-capability adopters of open source, while performance and strategic alignment dominate for the capable. This paper gives that claim a cost-based reading and an identification discipline. In a random-utility adoption model, a segment's capability is a unit implementation cost that scales the effort and support terms; the quantities acceptance research estimates correspond to marginal effects of the drivers on adoption probability. The claim that capability differs is imported from the typology. The cost primitive locates that difference in the cost-side drivers and states it as a claim about ratios.

Four results follow. First, ratios of marginal effects within a segment (support to performance, effort to performance) are free of both the error density and the segment's error scale. Their ordering across segments therefore identifies the capability ordering under any baseline and any scale heterogeneity, provided the technology's effort requirement and the support environment are common across segments. When support differs by segment, the support-to-performance ratio remains the protected carrier; the pair of ratios then identifies the support ordering as well. Second, levels of marginal effects confound structural sensitivity with the location of the segment on the adoption curve. The reversal condition is characterized exactly. On a declared configuration space, a naive level comparison misorders the capability ranking on 63 percent of random configurations under a probit link and 46 percent under a logit link; ratio comparisons are correct on all of them. Third, matching baselines eliminates the reversals when the error scale is common across segments and fails on 44 percent of the same space when it is not. The ratio route is therefore the recommendation and matching the fallback: the scale is the quantity the antecedent literature warns about, one that a single binary model does not identify. Fourth, in structural-equation practice, cross-group comparisons of standardized coefficients confound structure with variance heterogeneity. They differ across groups with identical structure on essentially every sampled configuration and reverse a true structural ordering on 17 percent of them; unstandardized comparisons on invariant metrics never do.

The difficulty of comparing groups in nonlinear models is established (Allison 1999; Mood 2010; Long & Mustillo 2021). The ratio cancellation of the first result is the textbook identification fact of discrete choice. This paper claims four things: the cost reading of segment dependence as a ratio claim; the exact reversal condition with its quantified shares; the within-segment cross-driver ratio as a carrier of the hypothesis that is invariant to both confounds at once; and the import of the discipline into acceptance research, where it is absent.

The practical prescription for the companion study: test segment dependence on ratio contrasts, with the support-to-performance ratio as the protected carrier; fall back on matched baselines only where a common error scale can be argued; report segment baselines alongside any probability-scale comparison; and never compare standardized coefficients across segments. None of this constitutes an empirical claim about the adoption data itself.

JEL: C25, C18, D22, O33, L86.

About this paper¶

Cite this paper¶

Fermigier, S. (2026). Why Driver Weights Differ: A Cost-Based Microfoundation for Segmented Technology Acceptance.
Draft working paper v0.3, Abilian Econ Lab.
https://econ.lab.abilian.com/papers/acceptance-structure/