Case Study 1: Credibility in Practice — The Workers' Compensation Experience Modification Factor
What this case is. A look at the single most widespread real-world application of credibility theory in all of insurance: the experience modification factor — the "X-mod" or "e-mod" — that the National Council on Compensation Insurance (NCCI) and the independent state rating bureaus calculate for millions of American employers. The X-mod is owned and explored in full in Chapter 22; here we study it purely as credibility made operational — a filed, regulated, century-old formula that decides, for every employer, exactly how much its own loss history should move its price. Every named institution and mechanism below is real; no specific statistic, rate, or mod value is asserted as fact — the illustrative numbers are labeled as constructed.
Background: the problem the X-mod solves
Workers' compensation is statutory coverage — benefits are set by law, not negotiated in the policy (Chapter 22) — so insurers cannot compete on what the coverage pays. They compete on price, and the price starts from a published manual rate (a loss cost, in the sense of §10.3) for each class code: a rate per \$100 of payroll for "clerical," for "welding and structural-steel erection," for "trucking," and so on. Two employers in the same class code therefore start from the same manual rate. But two welding shops with the same payroll are not equally safe. One runs a disciplined hot-work program, trains its crews, and returns injured workers to light duty; the other cuts corners and bleeds claims. If both pay the manual rate, the safe shop subsidizes the dangerous one — a textbook adverse-selection problem (Chapter 1), and a moral-hazard one too (Chapter 1), because there is no price reward for safety.
The experience modification factor exists to break that subsidy. It adjusts each employer's manual premium up or down based on that employer's own past loss experience relative to what is expected for its class and size. A mod of 1.00 is exactly average for the class. A mod below 1.00 (a credit mod) means the employer has done better than expected and pays less; a mod above 1.00 (a debit mod) means worse than expected and pays more. The mod is applied multiplicatively: manual premium × mod = the experience-rated premium. It is, in plain terms, the answer to the exact question this chapter is about — how much should this employer's own experience move its price? — and the answer is computed by a credibility formula.
The underwriting / insurance issue: credibility, formalized and filed
Here is what makes the X-mod such a perfect study for this chapter: it does not trust an employer's raw experience. It credibility-weights it, and it does so through a filed, approved formula with several features that map directly onto §10.5–10.7.
Size drives credibility. A tiny employer with a handful of payroll dollars may not even qualify for experience rating — below an eligibility threshold, its experience is too thin to be credible at all, and it simply pays the manual rate ($Z$ effectively 0, the class does everything). As the employer gets larger — more payroll, more expected claims — it qualifies, and its experience earns progressively more weight. The largest employers' mods are driven substantially by their own losses (high credibility); the smallest qualifying ones' mods barely budge from 1.00 (low credibility). This is the square-root-rule intuition of §10.5 made into policy: more exposure, more credibility, and the relationship is deliberately less than proportional so that a small employer is never whipsawed by one bad year.
Frequency counts more than severity. This is the subtle, brilliant part, and it is pure §10.1 thinking. The X-mod formula splits each loss into a primary portion (the first slice of every claim, up to a limit) and an excess portion (the part above that limit), and it gives the primary (frequency-driven) portion far more weight in the mod than the excess (severity-driven) portion. Why? Because — exactly as §10.1 argued — frequency is a more stable, more credible signal of an employer's underlying safety than severity is. Whether a given accident turns into a \$50,000 claim or a \$1,500,000 claim is heavily a matter of chance (which way the worker fell), but how often workers get hurt at all is a far better measure of how the shop is run. So the formula trusts the frequency signal and heavily discounts the severity signal — it caps how much any single catastrophic claim can spike the mod, precisely because a single severe loss is a low-credibility, high-variance event. An employer is not destroyed by one freak claim, and is not rewarded for merely getting lucky on the size of the claims it did have.
The X-mod is, in effect, the Bühlmann insight of §10.7 hard-wired into a rating plan: it leans on the part of the experience that is mostly signal (frequency) and shrinks the part that is mostly noise (severity), and it scales the whole thing by size so that credibility rises with the volume of experience. An underwriter who understands §10.5–10.7 understands why the X-mod is built the way it is — and an underwriter who does not will misread mods constantly.
What it shows
The X-mod demonstrates, at national scale, every load-bearing idea in this chapter:
- Credibility weighting is real and regulated. Millions of employers are priced every year by a formula whose entire job is to decide $Z$ — how much own-experience versus class. It is not a classroom abstraction; it is the working price of American workplace safety.
- Small samples are not trusted. Eligibility thresholds and the less-than-proportional credibility design encode the §10.5 lesson that a few claims do not justify a punitive (or generous) price.
- Frequency is the credible signal. The primary/excess split operationalizes §10.1: an employer's safety shows up most reliably in how often, not how big.
- The mod aligns incentives. Because better experience produces a credit mod, the X-mod turns safety into money — it is loss control (Chapter 9) with a price tag, pushing back on the moral hazard of indifference to injury (Chapter 1).
It also shows the limits the chapter insisted on. The mod is backward-looking: it prices an employer on its past three-or-so years (lagged, because recent claims are immature — §10.4), so a shop that has genuinely transformed its safety this year will not see the reward until the improved years roll into the calculation, and a shop that has just begun to slide will look fine for a while. The mod also cannot see causation — it does not know whether last year's spike was a freak or a symptom — which is exactly why an underwriter's qualitative read (the loss-control report, the hot-work program) still matters on top of the number.
Outcome
Experience rating via the X-mod has been a durable, century-old fixture of U.S. workers' compensation, and the machinery around it is entirely public: NCCI and the independent bureaus publish the Experience Rating Plan Manual, file it with state regulators, and recalculate mods annually from data carriers are required to report. The plan has been revised over the decades — the split point between primary and excess losses, in particular, has been adjusted (and at one point phased upward over several years) as loss costs and medical inflation changed what counted as a "small" claim — but the credibility architecture has held: size-scaled credibility, frequency weighted over severity, single large losses capped. Disputes are common and instructive — employers challenge the loss data feeding their mod, and a single mis-coded or over-reserved claim can move a mod and a premium materially — which is itself a §10.2/§10.4 lesson: the quality and maturity of the loss data feeding a credibility calculation matter as much as the formula.
Lesson
The transferable lesson is the chapter's thesis, proven at scale: credibility is not an academic nicety; it is how the industry decides, every day, how much a risk's own story is allowed to move its price — and the best designs trust frequency over severity, scale trust to the volume of experience, and refuse to let one large, low-credibility loss stampede the number. When you build Harbor Steel's workers'-comp price in Chapter 22, you will be applying this formula, and you will read its mod correctly only because you understand the credibility theory underneath it. And when you are tempted, anywhere in your career, to price an account on two or three of its own claims, remember that the largest credibility machine in insurance was built specifically to stop you from doing that.
Discussion questions
- The X-mod weights the primary (frequency-driven) portion of losses far more heavily than the excess (severity-driven) portion. Tie this directly to §10.1's claim that frequency and severity carry different amounts of signal about the underlying risk. Why is "how often" more credible than "how big"?
- A small employer suffers one serious claim and its mod rises only modestly; a much larger employer with the same single claim sees almost no mod movement at all. Using the square-root-rule intuition (§10.5), explain why size produces this difference — and why that is fair rather than a loophole.
- The mod is computed from an employer's experience lagged by a year or more, because recent claims are immature. Connect this to §10.4 (development). What does it imply about a company that has genuinely fixed its safety problem this year?
- The X-mod cannot see causation — whether a loss was a freak or a symptom. Where, then, does the underwriter's judgment (and the loss-control read of Chapter 9) add value on top of the credibility-weighted mod? Give a concrete example.
- An employer protests that its debit mod is "punishing one bad accident." Using primary/excess splitting and capping of large losses, explain why the mod is in fact designed to limit how much one severe claim can hurt — and what the protest reveals about a common misunderstanding of credibility.