Does postselection bias equally distort positive and negative thrust measurements, or does it preferentially amplify one direction?

# Postselection Bias and Thrust Measurement Directionality

This is an excellent question about whether selection effects introduce *symmetric* distortions or *preferentially amplify* signals in one direction—a crucial distinction for interpreting null or anomalous results.

The preprints show that postselection bias is fundamentally asymmetric and tends to amplify positive deviations over negative ones, though the mechanism varies by context.

The Core Asymmetry

In gravitational-wave tests of modified gravity, selection bias causes **overestimation of a signal when it is truly present, but *no* spurious signal when absent** [AG-2025.01-657]. This is the critical asymmetry: the bias has a directional preference. When searching for a scalar polarization mode in gravitational waves, the analysis preferentially inflates the measured amplitude if that mode genuinely exists—but doesn't hallucinate it from noise. This is not neutral distortion; it's a one-way lever.

The reason is statistical: event selection (e.g., by signal-to-noise ratio) naturally correlates with the direction of the actual signal. If your selection criterion favors high-SNR events, you're implicitly biasing toward measurements that point *upward* along your parameter of interest.

Detection Thresholds Break Symmetry

More directly relevant to thrust claims: the detection probability itself is asymmetric [AG-2024.04-481]. When a source sits near a detection threshold (set by false-alarm rate), sources nominally below threshold are more likely to be detected than sources above threshold are to be missed. This creates an asymmetric "leakage" around the threshold. For propulsion experiments reporting marginal positive thrust, this means a true null result could be more likely to leak upward across the threshold than a true small-positive result would leak downward.

Implications for Propulsion Claims

If a thrust-measurement protocol selects only events above some noise floor or SNR cutoff, postselection bias will preferentially amplify *positive* measurements. A genuinely null result is more likely to scatter upward into detection than downward into the noise floor, because the selection criterion couples to the magnitude, not the sign. Conversely, negative thrusts (or null results appearing negative due to noise) have no corresponding amplification—they're more likely to be filtered out entirely.

This directional bias is why careful statistical treatment—accounting for selection effects *in the likelihood itself*, not after the fact—matters [AG-2024.05-183]. The Bayesian approach, when applied properly, can recover true parameters even under selection, but only if the selection model is explicit in the inference.

---

Note: The provided preprints address selection bias in gravitational waves, particle physics, and cosmology, but don't directly analyze thrust-measurement protocols; a more specific paper on electromagnetic or mechanical propulsion measurement might refine these conclusions for that domain.