From a Claim to a Parameter
A research statement such as “the revised process reduces average handling time” sounds clear, but it does not yet specify what would be tested. A statistical hypothesis must make a claim about a population parameter, not merely about the observations collected in one sample. Translating the wording carefully is therefore an essential first step: identify the population, name the parameter, determine which values it can take, and express the claim and its benchmark in that notation.
This translation does not decide whether evidence is convincing. It sets out what the statistical question is. A sample average or sample proportion may later provide evidence about a population quantity, but it is not itself the quantity named in the hypotheses.
A Reliable Translation Process
A useful translation starts with the subject matter rather than with symbols. First define the population and the outcome being measured. Then choose a parameter that describes that outcome, and state its units. Only after those choices should the words “higher,” “lower,” or “different” be converted into mathematical comparisons.
Specify whose outcomes are being described and what is measured. A claim about all eligible policyholders is not automatically a claim about only the policyholders sampled.
Use a population mean such as \(\mu\), a population proportion such as \(p\), or a parameter defined for a comparison. State the units or what the parameter represents.
List the values the parameter can take under the model. For a proportion, \(0\leq p\leq 1\); for a difference of two proportions, \(-1\leq p_1-p_2\leq 1\).
Decide whether the claim describes a departure from a stated benchmark, a comparison with another population, or a specified population value. Label which statement is the claim being assessed.
Use symbols only after the verbal claim is understood. Check that the population, direction of comparison, units, and parameter all agree with the original statement.
The notation must describe the population quantity in the claim. If a sample of policyholders has an average claim amount of $1,240, that observed average is a statistic. A hypothesis about the long-run average claim amount concerns the population mean \(\mu\), not the observed sample average. Similarly, a sample fraction who renew is evidence about a population renewal proportion \(p\); the sample fraction is not \(p\).
The parameter space is more than a technical detail. It can expose an impossible interpretation, such as a proposed population probability of 1.08, or clarify what a difference means. A difference in proportions can be negative, zero, or positive, even though each proportion separately must lie between zero and one.
Worked Example: A Claim About Average Processing Time
A hypothetical benefits administrator sets a target of 15 minutes for mean processing time per completed application. A research team claims that a revised workflow brings the population mean below the target. Define the parameter, give its parameter space, and translate the claim and reference into hypotheses.
Method: define a population mean, then translate the directional claim. Let \(\mu\) be the mean processing time, in minutes, for all completed applications handled by the revised workflow under the conditions of interest. The outcomes are times, so the mean is positive; its parameter space is \(\mu>0\).
The benchmark is 15 minutes. The team’s claim “below the target” means a comparison of \(\mu\) with 15 in the downward direction. The benchmark statement is \(H_0:\mu=15\), and the research claim is \(H_A:\mu<15\).
The claim being evaluated is \(H_A\), not a claim about the average in a particular sample. For example, a sample mean of 14.6 minutes would be a statistic that could inform the question, but it would not change the definition of \(\mu\).
Common mistake avoided: Writing \(H_A:\bar{x}<15\) would make a hypothesis about the observed sample mean. The research statement concerns the population mean, so the parameter is \(\mu\). The units also match: both \(\mu\) and the benchmark are measured in minutes.
Check: If the population mean equals the stated benchmark, then \(\mu=15\), which satisfies the reference statement and not the strict “below 15” research claim. The translation preserves the direction and the benchmark.
Claims About Proportions
Many actuarial questions describe the fraction of a population with a specified outcome: a renewal, a particular claim type, or a successful completion. In such cases, define the event precisely and let \(p\) be its probability in the population of interest. Since \(p\) is a probability, its parameter space is \(0\leq p\leq 1\).
Care is needed with percentage wording. “More than 72%” corresponds to a proportion greater than \(0.72\), while “72 percentage points higher” is a statement about a difference between two proportions. A percentage change relative to a starting value is different again: it divides the change by the starting proportion. State which quantity is being claimed before writing symbols.
Worked Example: A Claim About a Renewal Proportion
A hypothetical insurer reviews a new reminder program. The research claim is that more than 72% of eligible policyholders offered the program will renew by the specified date. Define the parameter, specify its parameter space, and write the hypotheses. Explain why a sample renewal fraction is not the parameter.
Method: model the outcome as a population proportion. Define “success” as renewal by the specified date among eligible policyholders offered the program. Let \(p\) be the probability of this outcome for a policyholder in that population under the program. Because \(p\) is a probability, \(0\leq p\leq 1\).
The stated threshold is 72%, which as a proportion is \(0.72\). “More than” gives the research claim \(H_A:p>0.72\). The benchmark statement is \(H_0:p=0.72\). Thus the claim being assessed is that the population renewal probability exceeds 0.72.
Suppose a study observes 156 renewals among 200 eligible policyholders. The sample fraction is
The value \(0.78\) is a sample statistic, while \(p\) is the unknown population parameter. The observation is consistent with the direction of the research claim, but writing \(H_A:\hat{p}>0.72\) would not express a population hypothesis.
Common mistake avoided: Using 72 rather than \(0.72\) as the benchmark would put a probability outside its parameter space. Converting 72% to \(0.72\) before writing the hypothesis keeps the units and scale consistent.
Check: The statistic \(0.78\) lies in the proportion parameter space, and the benchmark \(0.72\) does as well. The hypotheses refer to the same defined event and population.
Claims Comparing Two Populations
A claim that one population has a lower rate than another must identify both populations and the order of subtraction. Let \(p_1\) and \(p_2\) be the event probabilities in populations 1 and 2. Defining \(\Delta=p_1-p_2\) makes the comparison explicit: \(\Delta<0\) means the probability in population 1 is lower. Since both component proportions are between zero and one, \(-1\leq\Delta\leq1\).
This definition prevents a common sign error. If the order is reversed to \(p_2-p_1\), the same verbal claim has the opposite inequality. Always write the definition of a difference next to the hypotheses, particularly when the population labels are not self-explanatory.
Worked Example: Comparing Two Claim-Filing Rates
A hypothetical insurer compares the share of customers filing at least one glass claim within a year. Population 1 consists of customers offered a revised warning notice; population 2 consists of customers offered the existing notice. The research claim is that the revised notice is associated with a lower filing proportion. Define the comparison parameter, its space, and the hypotheses.
Method: define a signed difference in population proportions. Let \(p_1\) be the one-year filing probability for customers offered the revised notice, and let \(p_2\) be the corresponding probability for customers offered the existing notice. Define
Each proportion lies between zero and one, so the difference has parameter space \(-1\leq\Delta\leq1\). The words “revised notice is associated with a lower filing proportion” mean \(p_1<p_2\). Subtracting \(p_2\) from both sides gives \(p_1-p_2<0\), or \(\Delta<0\).
The hypotheses are \(H_0:\Delta=0\) and \(H_A:\Delta<0\). The claim being assessed is the alternative statement that the revised-notice population has the lower proportion.
For a numerical sign check, suppose the population proportions under consideration are \(p_1=0.12\) and \(p_2=0.17\). Then \(\Delta=0.12-0.17=-0.05\), which is negative, as required by the claim. This check uses hypothetical values only to verify the direction of the definition.
Common mistake avoided: If the parameter were instead defined as \(p_2-p_1\), its value in the check would be \(0.17-0.12=0.05\), and the research claim would require a positive, not negative, difference. The error is not in the data; it is in changing the subtraction order without changing the inequality.
Check: Both proportions satisfy their individual parameter spaces, the difference is within \([-1,1]\), and a lower filing proportion in population 1 corresponds to a negative value of the defined \(\Delta\).
Claims About Levels, Changes, and Units
A statement about an absolute level should not be confused with one about a change. “The mean is 4 dollars higher than the benchmark” compares a mean \(\mu\) with a benchmark plus 4 dollars. “The mean increased by 4%” describes a relative change, whose denominator must be specified. A difference between means has the same units as the measurements; a ratio of means is unitless.
Likewise, “at least” and “at most” describe inclusive thresholds in ordinary language, whereas “more than” and “less than” describe strict comparisons. The exact way a threshold is assigned between the null and alternative is a matter of hypothesis formulation. At this stage, the practical priority is to define the parameter, preserve the meaning and direction of the claim, and distinguish the stated claim from its reference. Do not silently replace a claim about “at least 80%” with one about “more than 80%.”
| Claim wording | Parameter to define | Translation check |
|---|---|---|
| Mean payment is below $900 | Population mean payment \(\mu\) | Use dollars and compare \(\mu\) with 900 |
| Renewal probability exceeds 0.70 | Population renewal proportion \(p\) | Confirm the event and population behind \(p\) |
| Group A has a lower event rate than group B | For example, \(\Delta=p_A-p_B\) | Under this definition, lower in A means \(\Delta<0\) |
| Average duration differs from 12 days | Population mean duration \(\mu\) | “Differs” does not specify a direction |
Common Translation Errors
- Using a statistic instead of a parameter. A sample mean \(\bar{x}\) and sample proportion \(\hat{p}\) summarize observed data. A hypothesis about a population is written using the corresponding population parameter.
- Leaving the population vague. “The proportion is 60%” is incomplete unless the outcome and population are specified. A rate for new customers may differ from a rate for all customers.
- Reversing a difference. If \(\Delta=p_A-p_B\), then “A is lower than B” means \(\Delta<0\). Reversing the subtraction reverses the sign.
- Mixing scales. Convert percentages to proportions when using \(p\), and do not compare a dollar amount with a benchmark stated in thousands of dollars without converting units.
- Treating the research claim as automatically being the null. A claim can be the statement of interest while the null is a reference or benchmark. Label the claim explicitly rather than deciding from wording alone.
Check Your Understanding
For each situation, identify the parameter and its parameter space when relevant, then translate the stated claim into hypotheses. Answers emphasize the parameterization and direction.
- A hypothetical service center claims that the mean time to resolve a qualifying request is below 8 hours. Let \(\mu\) denote the population mean time. State the parameter space and hypotheses. (Answer: \(\mu>0\); \(H_0:\mu=8\), \(H_A:\mu<8\).)
- A study claims that more than 0.64 of eligible applicants complete an online application. Define \(p\) as the population completion probability. State the parameter space and hypotheses. (Answer: \(0\leq p\leq1\); \(H_0:p=0.64\), \(H_A:p>0.64\).)
- For a hypothetical comparison, let \(p_A\) be the proportion of Group A accounts with an overdue balance and \(p_B\) the corresponding proportion in Group B. A claim says Group A has the higher proportion. Define \(\Delta=p_A-p_B\) and write the hypotheses. (Answer: \(-1\leq\Delta\leq1\); \(H_0:\Delta=0\), \(H_A:\Delta>0\).)
- A product team claims that the mean annual service cost for a new plan is at least $30 lower than for the current plan. Let \(\mu_N\) and \(\mu_C\) be the respective population mean costs, in dollars, and define \(\Delta=\mu_N-\mu_C\). Translate the threshold claim and specify the parameter’s units. (Answer: The claim is \(\Delta\leq-30\) dollars; \(\Delta\) is a difference in mean costs measured in dollars.)
- A hypothetical program’s completion probability is claimed to be 8 percentage points higher for participants than for nonparticipants. Define \(p_P\) and \(p_N\) as the two population probabilities and \(\Delta=p_P-p_N\). State the claim in proportions and give the parameter space. (Answer: \(\Delta=0.08\) as the stated increase; \(-1\leq\Delta\leq1\).)