"Walking on eggshells": How expressing relationship insecurities perpetuates them.

The core claim is about a self-reinforcing cycle in close relationships over time. Testing it requires tracking the same relationships longitudinally and observing each link operating in sequence. Only Study 4 attempts this, with 37 friendship pairs — predominantly platonic, not romantic, despite the model being framed around romantic insecurity.

The most pervasive problem is same-source bias: in Studies 2A, 2B, and 3, every variable is reported by the same person in the same session, so a general disposition toward negativity could produce all the observed correlations. The expression-of-vulnerability and reflected-appraisals scales share near-identical items ("I have often reacted in ways that suggest I am sensitive about rejection" vs. "This person thinks I am sensitive to his/her opinions of me"), inflating correlations through item overlap rather than a causal process.

Study 5, the only experiment, uses a condition-blind confederate (a genuine strength), but the manipulation — telling participants after a ten-minute conversation that a stranger had read an unflattering description of them — introduces ego threat as an alternative explanation, and no manipulation check confirms participants believed the cover story. The paper does control for self-esteem and attachment anxiety, and Study 4 collects partner reports. But the cycle is never demonstrated operating in any actual relationship.

No study includes a power analysis or sample-size justification. Study 4's longitudinal model fits eight or more variables to 37 dyads. One link in the cycle (reflected appraisals to authenticity doubts at the second time point) is marginal, and the model's fit index (.086) exceeds the paper's stated threshold for reasonable fit (.08). A footnote reveals the full composite outcome did not produce the expected result; the paper switches to a subscale that did.

Studies 2A and 2B use cross-sectional Sobel tests to claim mediation — a method that cannot distinguish causal ordering from a shared underlying factor. Study 5's primary comparison (vulnerability condition vs. no-information control) falls short of conventional significance at the two-tailed level; the paper notes one-tailed tests would reach significance, but this appears post-hoc. The only conventionally significant experimental comparison is against a condition where participants were told the stranger had been informed they were "extremely insensitive to social cues" — an unflattering description that introduces its own confounds. No confidence intervals appear anywhere, and no corrections are made for the many tests across six studies.

The model describes four distinct cognitions in a causal chain, but when all four are measured by self-report from the same person in the same session — as in three of six studies — they may not be four distinct things. They may be four ways of asking one person whether they feel insecure. The expression and reflected-appraisals scales contain near-paraphrases, and the authenticity-doubts and perceived-rejection scales tap overlapping territory.

The paper controls for self-esteem and attachment anxiety, then claims the process operates "independently of individual differences." But rejection sensitivity — the most directly relevant trait, discussed at length in the literature review — is never measured. Neither are neuroticism, depression, or negative affectivity. Independence from two traits is not independence from individual differences generally.

The proposed egocentric-bias mechanism (spotlight effect, illusion of transparency) is imported from research on strangers without any direct test in this paper. Study 4's data actually show that partners do detect expressed vulnerability and that by the second time point, partners thought to see the participant as vulnerable report being less authentic — evidence that the process may not be purely intrapersonal. Study 5 tests a plausible analogue of one link, but between strangers with a two-item outcome, far from the model's domain of ongoing relationships and broad authenticity doubts.

The reasoning is internally coherent — given the assumptions, each step follows from the last. But the evidence never catches up to the theory. The full cycle is never tested in one population. The longitudinal test is underpowered, with one marginal link and a selectively reported outcome. The cross-sectional mediation analyses cannot distinguish the proposed causal chain from a single underlying factor. The experiment tests one link with a marginal result between strangers. The egocentric-bias mechanism — the theoretical centerpiece — is never measured or manipulated.

The paper claims this is a "normative" process operating in anyone regardless of personality; controlling for two individual-difference variables does not establish that. The paper acknowledges some gaps (the small longitudinal sample, the difference between the experimental setting and natural relationships) but treats them as minor caveats rather than fundamental limitations.

For a 2008 publication, the absence of pre-registration, open data, and open code is unremarkable. Survey items and experimental materials are reproduced in the text, and funding is disclosed. But several analytic choices are buried: the switch from the composite outcome to a subscale in the longitudinal model appears only in a footnote; the choice between one-tailed and two-tailed tests in Study 5 appears post-hoc; and the limitations discussion omits same-source bias across three studies, item overlap between scales, and the selective outcome reporting. The reflected-appraisals scale's low reliability (.61 at Time 1 in Study 4) is not discussed despite its implications for mediation and control analyses.

The citations are drawn from well-established research programs and used with care. But the egocentric-bias mechanism is built on studies conducted with strangers (the Gilovich group's work on the spotlight effect and illusion of transparency) and applied to close relationships with minimal justification. The paper underplays its own evidence that authenticity doubts may sometimes be accurate — Study 4 shows partners do notice vulnerability and adjust their behavior. The citation network is narrow, centered on the Murray/Holmes group and the authors' prior work, with Study 4's dataset shared with another Lemay & Clark publication from the same year. No literature on when vulnerability disclosure strengthens relationships — a well-documented phenomenon — is integrated into the model's development; it appears only as an afterthought in the Discussion.

Both funding sources are US government agencies (NIMH, NSF) with no commercial stake. This is basic research with no product, therapy, or commercial application under study. No conflicts of interest are identifiable.