Response to AOL's 2nd FR Comments

BLS response to August 1, 2007 AOL comments submitted to OMB.

A. Comments That Evaluate Whether The Proposed Collection Of

Information Is Necessary For The Proper Performance Of The

Functions Of The Agency, Including Whether The Information Will

Have Practical Utility.

AOL believes the survey exceeds the authority granted to the BLS by the Occupational

Safety and Health Act ("OSH Act"). Under OSHA recordkeeping standards, employers

in low-hazard industries are partially exempt from injury and illness recordkeeping.'

Through an appropriate rulemaking process, OSHA determined that there was little

practical utility in requiring low-hazard industries to keep injury and illness data. In

addition, agencies are required to avoid regulations that are inconsistent, incompatible, or duplicative with its other regulations or those of other federal agencies. As stated in the OSH Act's preamble:

"Congress intended.. . to minimize the recordkeeping burden on all employers,

not only small businesses. Exempting from routine injury and illness reporting

requirements those employers whose records are unlikely to be of significant

benefit to OSHA, or to the employers and their employees, serves this important

interest."

The BLS Survey is incompatible with OSHA regulations in that a number of large

exempt employers, otherwise exempt from OSHA recordkeeping requirements, are

chosen "with certainty" - thus effectively regulating that which is statutorily unregulated.

This burden on exempt employers is significant and contrary to Congressional intent. In

doing so, BLS is acting beyond the scope of the Survey authorization granted by the Act.

As noted above, 29 CFR 1904.42tb) notes that the exempt companies chosen to

participate in the Survey must be randomly selected and the BLS Survey and OSHA

regulations consistently cite a "sample selection process" or "random sampling," thereby

implying random selection. Yet the chosen method of stratification often results in a

limited number of large employers in a SIC/NAICS code in a particular state. Thus, to

obtain a statistically valid sample from this very small sample size, all the companies in

this particular category must be selected. These categories are called by BLS "certainty

cells" and, in Orwellian terminology, BLS notes that those companies are selected

"randomly with certainty" every year.

AOL responded to the March 19,2007 Federal Register notice inviting public comment

on the proposed revision of this same BLS Survey. In response to AOL's comments,

BLS admitted that "larger establishments tend to be sampled with higher probability or

possibly with certainty", though it should be noted that BLS has described in any number

of conversations that the companies are not "possibly" selected with certainty, but in fact

are selected "with certainty." BLS defended its practice by saying that this practice is

"more efficient than is random sampling", that the state agencies and/or the Bureau of

Economic Analysis and others find the data useful, and that BLS needs data "from good

actors such as AOL" to avoid the potentially erroneous data from others not as careful

with reporting. However, any rationalization - including agency efficiency or data utility

- does not change the fact that the Survey is required by DOL regulation to be random.

Surveying yearly at some large employers is beyond the scope of the BLS Survey

authorization. BLS notes in its comments:

"It is important to note that 'certainty establishments' are a ubiquitous feature of

establishment surveys, and the repeated surveying of an establishment over time

is not only a feature of certainty strata, but also of efficient surveys."

These "certainty establishments" are only "ubiquitous features" because of the BLS

sampling plan (stratification by state, then industry, then company size). BLS blames the

states for choosing certain industries for whom the states would like data; the states

blame federal BLS for choosing the particular establishments. By redesigning the survey

according to DOL authority, rather than starting with the states' desired company data

and working backwards, the BLS could ensure a truly random and statistically robust

sampling plan.

AOL recognizes that normally-exempt companies should be randomly selected for

inclusion in BLS's annual Survey. Our objection is the inclusion of certain of these

companies year after year on a non-random basis. With a potential sample population of

6.5 million companies, choosing the same company every year "with certainty" is not

random, regardless of the justification used to make this choice. Furthermore, AOL is

not advocating "simple random sampling" as the BLS's response to AOL's comments

seems to note. AOL recognizes that efficient sample design may not provide an

absolutely equal probability of any company being chosen. However, having a 100% (or

above 99%) probability of being chosen on a yearly basis is the root of the objection.

The regulatory process includes a Regulatory Impact Analysis ("RIA"). With the

"random selection with certainty," this RIA has been circumvented. With 85% of the

potential Survey pool normally exempt from OSHA recordkeeping, it would be expected

that the impact analysis of yearly selection would show a large impact.

The BLS collects data to produce a national estimate of all private industry and State estimates of private industry, and in many States, public sector. The BLS does not believe that data from employers with occupational injuries and illnesses are unlikely to be of benefit to the safety and health community. The following table gives the total recordable injury and illness rate for All U.S. and for Virginia in 2005:

Employment size class	US	VA
1-10	2.0	---
11-49	4.8	0.6
50-249	5.2	1.5
250-999	5.1	1.2
1,000+	5.2	0.3

AOL’s rate may not be equal to the average rate for establishments with 1000 or more employees, but their data are important in the computing of this rate.

As from BLS’ earlier response:

Public Law 91-596, 84 STAT. 1590, December 29, 1970, as amended through January 1, 2004 begins with the statement,

An Act - To assure safe and healthful working conditions for working men and women; by authorizing enforcement of the standards developed under the Act; by assisting and encouraging the States in their efforts to assure safe and healthful working conditions; by providing for research, information, education, and training in the field of occupational safety and health; and for other purposes.

Further in, Federal Register #: 66:5916-6135, published 01/19/2001, titled “Occupational Injury and Illness Recording and Reporting Requirements”, it is stated,

In order to further the purposes of this Act, the Secretary, in consultation with the Secretary of Health and Human Services, shall develop and maintain an effective program of collection, compilation, and analysis of occupational safety and health statistics. Such program may cover all employments whether or not subject to any other provisions of this Act but shall not cover employments excluded by section 4 of the Act. The Secretary shall compile accurate statistics on work injuries and illnesses which shall include all disabling, serious, or significant injuries and illnesses, whether or not involving loss of time from work, other than minor injuries requiring only first aid treatment and which do not involve medical treatment, loss of consciousness, restriction of work or motion, or transfer to another job. ...

Finally, the injury and illness records required by the OSHA recordkeeping rule are the source of the BLS-generated national statistics on workplace injuries and illnesses, as well as on the source, nature, and type of these injuries and illnesses. To obtain the data to develop national statistics, the BLS and participating State agencies conduct an annual survey of employers in almost all sectors of private industry. The BLS makes the aggregate survey results available both for research purposes and for public information. The BLS has published occupational safety and health statistics since 1971. These statistics chart the magnitude and nature of the occupational injuries and illnesses across the country. Congress, OSHA, and safety and health policy makers in Federal, State and local governments use the BLS statistics to make decisions concerning safety and health legislation, programs, and standards. Employers and employees use them to compare their own injury and illness experience with the performance of other establishments within their industry and in other industries. ...

An employer who receives an annual survey form from the Bureau of Labor Statistics must submit its annual injury and illness data to the BLS. The BLS conducts an annual survey of occupational injuries and illnesses under 29 CFR 1904.20-22 of the former rule (1904.42 of the final rule). Using a stratified sample, the BLS sends survey forms to randomly selected employers, including employers who, under Part 1904, would otherwise be exempt from the duty to keep the OSHA Log and Summary. These otherwise exempt employers are required to keep an annual record of the injuries and illnesses occurring among their employees that are recordable under Part 1904 if the BLS contacts them as part of the annual survey. At the end of the year, these employers must send the results of recordkeeping to the BLS. The BLS then tabulates the data and uses them to prepare national statistics on occupational injuries and illnesses. The BLS survey thus ensures that the injury and illness experience of employers otherwise exempted from the requirement to keep OSHA records -- such as employers with 10 or fewer employees in the previous year and employers in certain Standard Industrial Classification (SIC) codes -- is reflected in the national statistics.

The BLS is not acting beyond the scope of the survey authorization granted by the Act and is not incompatible with OSHA regulations. As seen from AOL’s comments, AOL disputes BLS’ use of stratified random sampling, unequal probabilities of sample selection and certainty establishments. The BLS has used stratified random sampling for the SOII since its inception in 1972. Under stratified random sampling, all possible establishments within scope of the survey are divided into separate groups or strata. Each stratum is assigned an allocation of the total sample, that is, each stratum is assigned a certain number of establishments that will be sampled. The allocation of sample under stratified random sampling is designed to maximize the amount of information available at a given fixed cost. This sampling approach is commonly utilized in surveys, including others in the BLS and other statistical agencies. In SOII, the strata are defined by State, industry, and establishment size. Once sample is allocated to strata, the establishments within each stratum are then sampled according to a simple random sample.

AOL states that “the chosen method of stratification often results in a limited number of large employers in a SIC/NAICS code in a particular state.” This statement gives the mistaken impression that BLS only samples large establishments to produce estimates for a particular SIC/NAICS. In fact, to be able to produce meaningful industry size class statistics, all industry/size class strata need to be sampled. To that end, the BLS draws samples that include small, medium and large establishments. There are strata where the number of establishments is not sufficient to meet the number allocated. In these instances, all of the establishments in these strata are chosen. This may not be AOL’s definition of random sampling, but it is stratified random sampling.

AOL accuses the BLS of “blaming” the States and vice-versa. The BLS and the participating States have a 50-50 relationship where the States contribute 50% of the money needed to run the program in their State and the BLS funds the other 50%. States have special interests in certain industries in their States. That is the reason for the concept of Target Estimation Industries (TEI) that was introduced into this program in the mid-1990’s, according to which States designate TEIs for which the States wish to have estimates. It would be foolish and wrong to force Kansas to produce estimates for shellfish fishing. BLS places no blame on Virginia if they wish to produce data for the industry in which AOL resides. And BLS is not to blame for randomly selecting establishments in strata – even in strata where there are not enough establishments to meet the allocation target.

The BLS does not agree with AOL’s statement, “that BLS needs data ‘from good

actors such as AOL’ to avoid the potentially erroneous data from others not as careful

with reporting”. The BLS does not classify sample employers into “good actors” and “others not as careful with reporting” nor does BLS even have information to make such a classification. BLS’ regional offices and participating States work hard to insure that all data are accurate.

AOL states, “By redesigning the survey according to DOL authority, rather than starting with the states' desired company data and working backwards, the BLS could ensure a truly random and statistically robust sampling plan.” First, the survey does not start with the “states’ desired company data.” The States do not identify the companies that are to be included in the survey. Rather, the States identify Targeted Estimation Industries for which they would like to have estimates. The BLS then draws a sample to produce these estimates in an efficient manner. Second, this “truly random” sampling method that AOL advocates would not allow a sufficient sample in the larger size classes and, unless the sample size was prohibitively large, may bypass the larger size classes entirely. What would result is a much more inferior product for the nation and the States to rely on for occupational safety and health.

AOL states that they do not want a 100% or 99% probability of being selected. There are times that strata have so few establishments, that even a 60% or a 55% probability of selection would still insure the selection of the establishments in the strata.

The BLS does not believe it has circumvented any regulation. Stratified random sampling is designed to maximize the amount of information available at a given fixed cost. This sampling approach is commonly utilized in surveys, including others in the BLS and other statistical agencies.

B. Comments That Evaluate The Accuracy Of The Agency's Estimate Of

The Burden Of The Proposed Collection Of Information, Including

The Validity Of The Methodology And Assumptions Used.

AOL believes the proposed data collection incorrectly estimates the burden for OSHA exempt employers. BLS drastically underestimates the recording burden and time it

takes to record the required data for normally-exempt employers. Because over 75% of

the Survey sample population are companies in low hazard industries (and thus typically

exempt from OSHA recordkeeping), these companies typically have little or no corporate

infrastructure to support the BLS Survey request - unlike non-exempt companies, who

have both an administrative process and safety staff in place to facilitate this

recordkeeping.

The OMB time estimates for completion noted on the OSHA forms used in the BLS

survey were prepared for, and derived from, this non-exempt (regulated) company

population, who must merely transfer prerecorded data to the OSHA form. Although it is

recognized that the BLS does provide burden estimates that are higher for normally

exempt employers than for non-exempt employers, these estimates are derived, in

circular logic, from the original OSHA forms prepared for non-exempt employers. In

December prior to the year of data collection, the BLS pre-notifies sampled employers

who are normally exempt from maintaining OSHA forms. The OSHA forms are sent to

these employers along with methods for the employers to receive the complete detailed

recordkeeping instructions. Included in the burden hours estimate for these normally

exempt employers is an average of 1 hour 21 minutes to keep the OSHA forms during the

year. Yet, this estimate itself is derived from the OSHA 300A log noting that each new

entry on the log requires 14 minutes on average. Exempt companies do not have preexisting familiarity with the 300A log - indeed, they have no familiarity with the arcane rules of what is or is not an OSHA recordable incident, no staff familiar with the vagaries of the little boxes on the log, no recordkeeping systems nor procedures set up to funnel the incident reports to the log-keeper, etc. Thus, even with pre-notification, the time burden for becoming familiar with this entirely new program is far greater than the time noted on the OSHA forms.

While the BLS Form 9300 indicates that completing the records for each site could take

anywhere from 10 minutes to 5 hours, AOL has found an average time burden across

nine sites of varying size to be 16.19 hours of time per site for completion the Survey.

This is only an average, and yet it is significantly higher than the time burden estimated

on BLS Form 9300. The BLS form should provide a burden estimate that more

realistically reflects the actual burden imposed on usually exempt employers, particularly

given that the majority of BLS Survey respondents are usually-exempt employers

(approximately 75% of respondents).

The BLS Supporting Statement also claims that "[no] capital costs or startup costs are

incurred in the recordkeeping or reporting functions of this survey." As noted above, due

to the lack of infrastructure, AOL expended many internal staff hours and hired a

consultant in order to facilitate completion of these forms. These startup costs were

indeed significant. The costs of generating, maintaining, and disclosing the information

requested are not only direct burden hours on AOL staff, but also include a yearly fee of

approximately $10,000 in externally paid costs.

The burden placed on typically-exempt employers by the Survey in the past will only

increase with this new proposal. Currently, BLS only collects case and demographic data

for injury cases with days away from work, but will test collection of case and

demographic data for injury and illness cases that require only days of job transfer or

restriction beginning in 2008 and continuing in 2009 if successful. Typical disability management systems in low-hazard industries usually show return-to-work dates only.

Adding the functionality to track restricted work days or job transfer is indeed an extra

cost and burden as yet unquantified. The same argument can be made as above - though

the data may be useful, that utility should not provide automatic authorization for the

BLS to collect it without a careful impact analysis. This new recordkeeping requirement

will result in a substantial time burden increase over that which exists today.

AOL states that the average time burden across nine sites was 16.19 hours. It would be interesting and enlightening for AOL to break down these hours into specific tasks and establishments, so BLS could better understand this time estimate. It is important to note that AOL is in a very low hazard industry and some of the sampled establishments are small. So, the number of injury and illness cases for which AOL must report should be zero or small even in AOL’s large establishments. AOL’s very large average time estimate is inconsistent with the reality of AOL’s limited reporting burden for a small number of cases.

AOL states that there were significant start-up costs, including hiring a consultant. The BLS believes that most establishments do not need to hire consultants to fill out the injury and illness forms. It is also BLS’ belief that many large exempt firms already have the infrastructure in place to record injuries and illnesses, if for no other reason than to handle Workers’ Compensation (WC) claims and recording.

AOL states that the workload will become larger with the addition of cases with days of job transfer or restriction only (DJTR). AOL also states that “adding the functionality to track restricted work days or job transfer is indeed an extra cost and burden as yet unquantified.” In fact, all employers, nonexempt and exempt, currently track DJTR cases on the forms, so there is no additional recordkeeping burden. Further, the BLS believes that the OSHA forms with the “little boxes” that are currently used readily explain to the employer to mark the most severe injury/illness category that pertains to the case. If the fatality box is not checked and the DAFW box is not checked, the forms design leads the employer to the DJTR box or the other recordable case box.

The proposal for expanded data collection is that BLS will collect case and demographic information for DJTR cases, just as it does now for cases with days away from work (DAFW). BLS plans to maintain the upper limit on the number of cases for which each employer must submit case forms at 15 cases – days away from work cases (DAFW) and days of job transfer or restriction cases (DJTR) combined. This limit was reduced from 30 cases in previous years.

The BLS employed a new sample allocation procedure starting with survey year 2006 that changed from targeting a specific variance for each industry - and thereby getting larger numbers of exempt employers in the sample – to more heavily allocating sample to industries with higher number of cases. The new sample allocation should lessen the burden on smaller exempt establishments, while the limit of 15 cases should lessen the burden on larger establishments.

C. Comments That Enhance The Quality, Utility And Clarity Of The

Information To Be Collected.

The Supporting Statement notes that "The purpose of the Act, as stated in Section 2(b), is

to assure, as far as possible, every working man and woman in the nation safe and

healthful working conditions. The BLS Survey provides the Nation's primary indicator

of the progress towards achieving this goal. The survey measures the overall rate of occurrence of work injuries and illnesses by industry." The Supporting Statement also

notes that "[s]urvey data are used to assess the Nation's progress in improving the safety

and health of America's work places; to prioritize scare(BLS note: misspelling “scarce”) Federal and State resources; to guide the development of injury and illness prevention strategies; and to support OSHA and State safety and health standards and research."

The data obtained will not be an accurate indicator of the working conditions of "every

working man and woman" if the Survey contains a certain group of the same employees

year after year. Thus, to increase the quality and utility of the data, the BLS should reevaluate its sample stratification design to ensure true random sampling. The design as

presently set forth backs BLS into a mathematical corner, with a very small sample

population in the large company category, thus causing BLS to choose the same

companies year after year. Random selection of companies, as noted above, should be

mathematically and statistically random.

The previous paragraph is certainly not the belief of BLS or of the community of survey statisticians. The statement could logically be extended to the extreme to argue that if the survey was a census, and therefore collected from the same group of employers each year, the results would not be an accurate indicator. In fact, the result would be a census of all injuries and illnesses and could not be any more accurate. Stratifying across State, industry, and size class yields not only a more efficient sample but a much more representative sample of the country’s occupational injury and illness experience. It is important to reiterate that the survey samples from all industries and size classes and does not single out certain large establishments. But, without the largest strata represented, the survey would indeed provide a very inaccurate indicator. Finally, the BLS believes that the sample chosen for this survey is mathematically and statistically random.

III. CONCLUSION

The Paperwork Reduction Act Notice of March 19,200'7 fails to accurately quantify or

characterize the administrative and financial burdens on OSHA-exempt employers like

AOL with regard to the proposed data collection. The proposed action will impose a

burden on employers who are typically exempt from OSHA injury and illness reporting

that is unnecessary to the goals of the BLS or the DOL as established by Congress.

We urge OMB to require BLS to revise the Survey design, as well as the time burden

estimate associated with completion of the Survey by the largest proportion of typically

OSHA-exempt employers.

BLS believes that it has designed a statistically efficient and valid survey that produces accurate estimates for workplace injuries and illness for States, industries and employment size classes. The burden that is imposed on respondents is consistent with the need for an efficient and accurate survey design and with the BLS mandate to develop and maintain an effective program of national and State statistics. BLS continues to believe that its average time burden estimates are accurate, even if a particular respondent such as AOL reports an atypical experience.

In closing, AOL stated in the meeting with the Solicitor’s Office and the BLS, that having their establishment in [State A] surveyed once in the last 10 years and their establishment in [State B] selected twice in the last 10 years does not bother them. What bothers them is their establishment that gets selected each year. This viewpoint is inconsistent with their complaint about the burden imposed on normally exempt establishments. It would seem that hiring contractors for the establishments that get selected infrequently would be more costly than hiring a contractor, and then learning the duties needed to fill out the forms, in establishments that will maintain records every year.