Yesterday, Corp Fin issued new Guidance on Calculation of Pay Ratio Disclosure regarding the use of statistical sampling in connection with the pay-ratio disclosure requirement, which mandates public company disclosure of specified pay-ratio information, beginning with the upcoming 2018 proxy season. The new guidance provides a fairly expansive reading of the use of reasonable estimates, statistical sampling and other reasonable methods. But prepare yourself, it also uses terms such as “multimodal,” “gamma distribution” and, my favorite, “lognormal,” surely all firsts for this PubCo blog. (Wikipedia defines a “lognormal”distribution as “a continuous probability distribution of a random variable whose logarithm is normally distributed.” Does that help?) Whether or not you are mystified by some of the terminology (as am I), it is clear that the leitmotif (take that, statisticians) of the new guidance is that you can use or combine any number of different methodologies and estimates so long as they are all reasonable and appropriate under your particular facts and circumstances.

As discussed in this updated PubCo post (be sure to look at the revised version), the SEC also issued contemporaneous interpretive guidance regarding pay-ratio disclosure, which addressed three topics: company reliance on reasonable estimates, the use of existing internal records to determine the median employee and non-U.S. employees, and, most interesting, the use of other recognized tests and criteria (such as published IRS guidance) to determine employee/independent contractor status. For a more complete discussion of the pay-ratio rule, see our Cooley Alert, SEC Adopts Final Pay-Ratio Rule.

As you probably recall, the Dodd-Frank pay-ratio provision and related SEC rule require disclosure, in a wide range of SEC filings, of the ratio of the median of the annual total compensation of all employees of the company to the annual total compensation of the CEO. In light of substantial concern expressed regarding the difficulty and cost of implementing the rule, the SEC attempted to provide flexibility designed to balance the desire to reduce cost and burden with the need to maintain the intended benefits of the rule.

The introduction to the Corp Fin guidance sets the stage:

“In the adopting release for the pay ratio rule, the Commission expressly sought to provide flexibility to registrants by not specifying the ‘other reasonable methods’ that may be appropriate, allowing each registrant to determine the method that best suits its own facts and circumstances. The rule also provides for flexibility in setting the parameters of the statistical sampling method. In the adopting release, the Commission explained that registrants are permitted to make determinations based on their facts and circumstances and declined to specify requirements for statistical sampling, such as appropriate sample sizes, confidence levels, or other requirements, to avoid unduly constraining registrants from developing the most appropriate methodology.

“While providing broad flexibility, the Commission made clear that registrants must determine their own median and may not use industry estimates, such as the employee earnings estimates provided by the Bureau of Labor Statistics, as a substitute for determining their median. Additionally, the Commission clarified that a registrant seeking to use a lognormal distribution assumption in its statistical analysis can do so if it determines that the use of the assumption is appropriate given its compensation distributions.”

The new guidance from Corp Fin addresses a number of issues in connection in connection with the use of statistical sampling methodologies:

Combining estimates, sampling and other methodologies. Companies may combine the use of reasonable estimates with the use of statistical sampling or other reasonable methodologies, including by using different methodologies for different geographic or business units. To support that conclusion, Corp Fin relies on the use of “and/or” in Instruction 4.2 to Item 402(u), which provides that, in determining the median employee, the company “‘may use its employee population or statistical sampling and/or other reasonable methods.’ The use of ‘and/or’ in the Instruction indicates that a registrant is permitted to use statistical sampling, other reasonable methods or a combination of statistical sampling and other reasonable methods.” Moreover, the staff emphasizes that, by not specifying which “other reasonable methods” were appropriate, the SEC intended “to allow each registrant the flexibility to determine the method that best suits its own facts and circumstances.”

Examples of sampling and combined sampling. The staff emphasized that companies are permitted to use a combination of sampling methods. In particular, the staff notes that Instruction 4 to Item 402(u) does not limit the methods of sampling that may be used, providing only that companies “must use reasonable methods and make reasonable estimates.” In addition, the adopting release expressly allows the use of more than one statistical sampling approach in determining reasonable estimates of the median for companies with multiple business lines or geographical units. The staff also highlights that the release advises that “all statistical sampling approaches should draw observations from each business or geographical unit with a reasonable assumption on each unit’s compensation distribution and infer the registrant’s overall median based on the observations drawn.”

Corp Fin offers the following as examples of possible sampling methods:

  • “simple random sampling (drawing at random a certain number or proportion of employees from the entire employee population);
  • stratified sampling (dividing the employee population into strata, e.g., based on location, business unit, type of employee, collective bargaining agreement, or functional role and sampling within each strata);
  • cluster sampling (dividing the employee population into clusters based on some criterion, drawing a subset of clusters, and sampling observations within appropriately selected clusters);
  • cluster sampling may be conducted in one stage or multiple stages); and
  • systematic sampling (the sample is drawn according to a random starting point and a fixed sampling interval, every nth employee is drawn from a listing of employees sorted on the basis of some criterion).”

Examples of the use of reasonable estimates. Reasonable estimates can be used to identify the median employee and to calculate annual total compensation (or any elements of it) for employees other than the CEO. Corp Fin provides the following as examples of appropriate use of reasonable estimates (depending on the facts and circumstances, of course):

  • “analyzing the composition of the company’s workforce (by geographic unit, business unit, employee type);
  • characterizing the statistical distribution of compensation of the company’s employees and its parameters (e.g., a lognormal, beta, gamma or another distribution, or a mixture of distributions—for example a mixture of two normal or lognormal distributions yielding a bimodal distribution);
  • calculating a consistent measure of compensation and annual total compensation or elements of the annual total compensation of the median employee;
  • evaluating the likelihood of significant changes in employee compensation from year to year;
  • identifying the median employee;
  • identifying multiple employees around the middle of the compensation spectrum; and
  • using the mid-point of a compensation range to estimate compensation.”

Examples of other reasonable methodologies. To avoid limiting flexibility, the adopting release did not provide any examples of acceptable other reasonable methodologies. But, the staff emphasizes, as long as they are reasonable, companies can use any method or combinations of methods. The staff provides the following as examples of common statistical techniques and methodologies that companies may consider:

  • “making one or more distributional assumptions, such as assuming a lognormal or another distribution provided that the company has determined that the use of the assumption is appropriate given its own compensation distributions;
  • reasonable methods of imputing or correcting missing values; and
  • reasonable methods of addressing extreme observations, such as outliers.”

Hypothetical examples of the use of reasonable estimates, statistical sampling and other reasonable methodologies. Finally, the staff has provided the three hypothetical examples below to illustrate the use of reasonable estimates, statistical sampling and other reasonable methods. The examples are intended to elucidate the principles that a company may consider in light of the company’s specific facts and circumstances. In that regard, the staff cautions that some of the techniques below may be more suitable for larger companies with more complex workforces and that, for smaller companies or less complex situations, it may be more appropriate to use a simpler approach. The staff makes clear that the examples “are not meant to suggest that registrants follow any particular approach.”

Company A has employees in the U.S. and outside the U.S. within three business units and 21 geographic units, covered by multiple payroll systems.

  • One approach would be for the company to perform sampling from each of the three business units. In obtaining samples of compensation data from each of the three business units, the company selects samples from the geographic locations whose employee pay is generally representative of employee pay within the entire business unit.”

Company B has a global workforce with employees concentrated in the following geographic units: North America, China, Europe, and Latin America units.

  • The company may use a combination of statistical sampling and other methods to identify the median.
  • Where statistical sampling is used, the sampling method may be chosen so as to be reasonably representative of the employee population, based on the company’s knowledge of the workforce distribution across jurisdictions, composition of full-time and part-time employees, distribution of employees among typical occupations, and the company’s pay structures for typical occupations.
  • Within the North America geographic unit, the company employs mostly management and administrative employees at headquarters and a workforce consisting mostly of sales employees in 25 other cities. The company identifies the most common occupations of employees working at headquarters and draws a stratified random sample of headquarters employees other than the PEO in those occupations. Almost all employees outside headquarters are sales employees. Based on its understanding of employee pay outside headquarters, the company identifies three cities in which the distribution of employee pay and full-time and part-time employees is reasonably representative of the distribution of pay of employees outside headquarters. In those cities, the company randomly selects stores, from which a random sample of sales employees is drawn.
  • For employees in the Europe geographic unit, the company draws a stratified random sample of employees in typical occupations identified based on the company’s knowledge of its workforce and pay structure. Employees in the sample include managers, administrative personnel, service employees, and sales staff.
  • For the China geographic unit, the company uses a sample of full-time and part-time employees reasonably believed to be around the middle of the pay scale.
  • For the Latin America geographic unit, the information is drawn under a distribution assumption.
    • Based on the understanding of pay practices and workforce composition, employee pay in the Latin America unit is estimated to follow a lognormal distribution.
    • For example, the company may use reasonable estimates provided by regional managers to determine distribution parameters. Where pay ranges were considered, the mid-point of the pay range is used.
  • To identify the median employee, the company combines information from the North America, China, Europe, and Latin America geographic units, obtained as described above.”

“Company C has employees in the U.S. and Asia.

  • Based on the company’s information about its workforce composition and compensation policies, the company reasonably believes the distribution of employee compensation to be multimodal and approximately characterized as a mixture of lognormal distributions, weighted based on estimated workforce composition. The median may be identified based on the resulting distribution mixture.
  • As an example, the company may identify four main cohorts of workers: full-time employees in the U.S.; part-time employees in the U.S.; full-time employees in Asia; and part-time employees in Asia.
    • For the U.S. employees, distribution assumptions are based on data regarding pay levels and hours of a typical full-time and part-time employee at the company.
    • For international workers, distribution parameters are based on reasonable estimates of a typical full-time and part-time employee’s pay provided by regional managers.”