How Scoring Works

Our data sources

Every score on The Respect Index is derived from three categories of input. The primary source is confirmed candidate reports — structured submissions from job seekers describing their experience with a specific company's hiring process. Confirmation is done by email; confirmed reports carry five times the weight of unverified ones. The second source is public job posting data pulled from applicant tracking system APIs (Greenhouse, Lever, Ashby, Workday, and others), which feeds ghost posting detection and repost tracking. The third source is state WARN Act layoff filings. WARN data for 15+ states is aggregated via the Big Local News warn-scraper project (Stanford Computational Journalism Lab, Apache 2.0 license), supplemented by direct downloads from state labor agencies for California, Texas, and New Jersey. All WARN filings carry full attribution to the U.S. Department of Labor. WARN data is informational only and does not feed into hiring scores.

The four sub-scores

The Process Score is a weighted composite of four dimensions, each calculated from specific report questions. Communication (30%) measures how proactively the company kept candidates informed between stages and whether it responded to direct follow-up. Transparency (25%) measures whether the company gave accurate timeline estimates, whether the role matched its description, and whether rejection decisions were communicated with specificity. Efficiency (25%) captures the number of interview rounds and hours of unpaid work required — structured tests, case studies, or take-home assignments. Respect (20%) is drawn from a direct 1–5 rating of the overall candidate experience and, where applicable, whether a compensation offer matched the stated range.

Sparse scoring

Not every candidate answers every question. A respondent who had no take-home assignment cannot meaningfully rate test hours. A respondent who was rejected before the offer stage cannot rate offer accuracy. Rather than penalizing missing answers with a zero, The Respect Index uses sparse scoring: unanswered questions are excluded from the calculation entirely, and the weights of the remaining answered questions renormalize to fill the gap. A report that answers only Communication and Transparency questions still contributes fully to those dimensions. This means partial reports are always better than no report — every truthful answer improves the data.

Correcting for negativity bias

We openly acknowledge that candidates who had a poor experience are more likely to submit a report than those who had a neutral or positive one. We do not pretend this bias does not exist. Instead, we apply four structural corrections. First, scores are expressed as percentiles relative to all tracked companies, so a platform-wide negative skew affects every company equally and cancels out in comparative ranking. Second, Bayesian smoothing pulls companies with fewer reports toward the platform mean — a company with five reports cannot be pinned at an extreme outlier position; as reports accumulate, the score moves toward its true value. Third, confirmed reports carry five times the weight of unverified ones, and reports from candidates who received offers carry additional weight, since they represent the full arc of the hiring experience. Fourth, the scoring system weights factual, observable questions more heavily than feeling-based ratings — did the company respond, was a timeline given, how many rounds were required — because factual recall is more stable across mood and time than subjective sentiment.

Recency weighting

Hiring cultures change. A company that ghosted candidates two years ago may have rebuilt its recruiting team and now communicates well. To ensure scores reflect current behavior rather than legacy reputation, each report is weighted by when it was submitted. Reports from the last 90 days receive full weight (100%). Reports submitted within the past year but older than 90 days receive 75% weight. Reports older than one year receive 50% weight. Reports older than 36 months are archived and excluded from scoring entirely. This means companies can genuinely improve their scores by changing their practices — and that improvement shows up in the data within months, not years.

Thresholds and confidence

Scores are not displayed publicly until a company has accumulated at least five confirmed reports. Below that threshold, the platform shows a message indicating insufficient data rather than a potentially misleading early-data score. Once the five-report threshold is met, a confidence level is assigned and displayed alongside the score. Companies with 5 to 9 confirmed reports are labeled "Early data," indicating the score is real but may shift meaningfully as more reports arrive. Companies with 10 to 24 confirmed reports are labeled "Medium confidence." Companies with 25 or more confirmed reports are labeled "High confidence," meaning the score has stabilized and additional reports are unlikely to change it substantially.

Accounts and anonymity

Reports on The Respect Index never require an account and never will. If you choose to create a free account to use community features, the account is pseudonymous: you select a handle rather than your real name. Community forum participation requires an account; submitting a report does not. The two activities are kept strictly separate by design. If a report is submitted while a user is signed in, a salted one-way hash links the session to the report so the user can later update the outcome (for example, from "in process" to "rejected" or "offer accepted"). This hash cannot be reversed by any party, including us. No user identifier, email, or session data is stored on the report record itself. The hash-to-report mapping is encrypted server-side and is not accessible via any public API or client-side query.

We grade companies and processes, not individuals

The Respect Index grades companies and processes, never individuals. Reports and posts naming individual people are not accepted. All scoring, grading, and public data on this platform describes company-level behavior — hiring practices, process quality, ghost rates, and separation handling — not the conduct of any specific recruiter, hiring manager, or employee. This is both a design principle and a moderation rule. Individual names (even in positive contexts) are stripped from submissions and rejected by our forum moderation system.

What companies cannot do

Companies cannot pay to change, suppress, or improve their scores. The Respect Index accepts no payments from tracked companies and has no commercial relationship that would influence score display. Companies cannot request the removal of a report simply because they disagree with its rating — a negative experience, honestly reported, is not grounds for removal. Companies may formally dispute a specific report only on four grounds: documented factual inaccuracy, a clear violation of community guidelines, no verifiable professional relationship between the reporter and the company, or mistaken identity. All disputes are reviewed against the original submission and resolved based on evidence, not complaint volume. The dispute process is documented at the Dispute Process page.

Questions about methodology or data quality? hello@therespectindex.com. See also: How It Works · Dispute Process · Our Independence Charter