Privacy tool osha comments

March 8, 2014
To: The Department of Labor, Occupational Safety and Health Administration
Re: Proposed Rule: Improve Tracking of Workplace Injuries and Illnesses, Docket No. OSHA–
2013–0023
From: Micah Altman, Director of Research, MIT Libraries;
Non Resident Senior Fellow, Brookings Institution
David O’Brien, Project Manager, Berkman Center for Internet & Society, Harvard U.
Alexandra Wood, Fellow, Berkman Center for Internet & Society, Harvard U.
on behalf of Privacy Tools for Sharing Research Data Project, Harvard U.

We appreciate the opportunity to comment on the proposed rule for the electronic submission of
workplace injury and illness records that appeared in 78 Federal Register 67254 (November 8,
2013). These comments address the privacy risks associated with the proposed public disclosure
of such records. Our perspective is informed by substantial advances in privacy science that have
been made in the computer science literature and by recent research conducted by the members
of the Privacy Tools for Sharing Research Data project at Harvard University.1
As a general matter, we are proponents of transparency and open access, and we support the goal
of making more workplace injury and illness data publicly available. We and our colleagues
have previously published on the benefits to researchers and the public of wider data access.2 We
are also acutely aware of the challenges related to confidentiality that arise when collecting,
analyzing, and sharing data pertaining to individuals.3 In our research, we have studied various

1

The Privacy Tools for Sharing Research Data project is a National Science Foundation funded collaboration at
Harvard University involving the Center for Research on Computation and Society, the Institute for Quantitative
Social Science, the Berkman Center for Internet & Society, and the Data Privacy Lab. More information about the
project can be found at http://privacytools.seas.harvard.edu/.
2
See, e.g., Harvard Open Access Project, Berkman Center for Internet & Society,
http://cyber.law.harvard.edu/research/hoap .
A representative list of publications related to the Privacy Tools for Sharing Research Data project is available at:
http://privacytools.seas.harvard.edu/publications. King, Gary. "Replication, replication." PS: Political Science &
Politics 28.03 (1995): 444-452 ; Altman, Micah, and Michael McDonald. "THE PROMISE AND PERILS OF
COMPUTERS IN REDISTRICTING." Duke Journal of Constitutional Law & Public Policy 5 (2010). ; Altman, M.,
& Mcdonald, M. P. (2014). Public Participation GIS : The Case of Redistricting. Proceedings of the 47th Annual
Hawaii International Conference on System Sciences. Computer Society Press (IEEE); Altman, M., Mann, T. E.,
McDonald, M. P., & Ornstein, N. J. (2010). Principles for Transparency and Public Participation in Redistricting.
Brooking Institute.
3
See, e.g., Latanya Sweeney, “Matching Known Patients to Health Records in Washington State Data,” Data
Privacy Lab, IQSS, Harvard University (2013); Latanya Sweeney, Akua Abu, Julia Winn, “Identifying Participants
in the Human Genome Project by Name,” Data Privacy Lab, IQSS, Harvard University (2013); Amitai Ziv, “Israel's
‘Anonymous’ Statistics Surveys Aren't So Anonymous,” Haaretz (Jan. 7, 2013),

1

approaches to sharing sensitive data, and we strongly believe that a sophisticated approach to
data disclosure is needed in order to ensure both privacy and utility. As numerous reports by the
National Research Council have shown, naïve treatment of information confidentiality and
security has become a major stumbling block to efficient access to and use of research data.4
These comments respond to the following question raised in the Proposed Rule: “What
additional steps, if any, should the Agency take to protect employee privacy interests?” We argue
that OSHA’s approach towards “personally identifying information” is too narrow and that
releasing microdata (individual-level data) records exposes employees to re-identification risks.
By citing examples from the literature, we illustrate how an employee might be identified within
microdata records, even after personally identifying information, such as name, address, date of
birth, and gender, has been removed. After identifying some of the risks associated with the
proposed approach, we discuss data sharing models that provide stronger privacy protections and
make recommendations for a more nuanced approach to releasing workplace injury and illness
records.
Benefits from public data availability
Transparency is a fundamental principle of democratic governance, and public access should be
the “default setting” for information collected by government. Regular reporting and wider
dissemination of workplace injury and illness records will require only marginal additional effort
because businesses already compile these records under OSHA regulations, this information is
considered a public record, and it can easily be submitted electronically.
Routinizing the collection and dissemination of injury data is also better economic policy. It is
well established that when the cost of monitoring incidents is low, more regular monitoring and
gradual sanctions increase social welfare benefits.5 Furthermore, the costs of occupational injury
in the United States are at minimum tens of billions of dollars annually.6 Most of these costs are
not borne by the firms in which injuries occur, or by insurers, but are instead imposed on the

http://www.haaretz.com/news/national/israel-s-anonymous-statistics-surveys-aren-t-so-anonymous-1.492256
National Research Council, Expanding Access to Research Data: Reconciling Risks and Opportunities
(Washington: National Academies Press, 2005); National Research Council, Putting People on the Map: Protecting
Confidentiality with Linked Socialspatial Data (Washington: National Academies Press, 2007); National Research
Council, Beyond the HIPAA Privacy Rule: Enhancing Privacy, Improving Health through Research (Washington:
National Academies Press, 2009); National Research Council, Conducting Biosocial Surveys: Collecting, Storing,
Accessing, and Protecting Biospecimens and Biodata (Washington: National Academies Press, 2010).
5
A. Mitchell Polinsky and Steven Shavell, “The Economic Theory of Public Enforcement of Law,” Journal of
Economic Literature 38 (2000): 45-76; A. Mitchell Polinsky and Steven Shavell, “The Theory of Public
Enforcement of Law, in vol. 1 of Handbook of Law and Economics, ed. A. Mitchell Polinsky and Steven Shavell
(Amsterdam: Elsevier, 2007).
6
J. Paul Leigh, “Economic Burden of Occupational Injury and Illness in the United States,” Milbank Quarterly 89.4
(2011): 728-772.
4

2

individual and on the societal safety net.7 For these reasons, reductions in injury brought about
through better detection and changes in individual and firm behavior have the potential to yield
substantial benefits to individuals and to the economy.
Data collected from people and institutions have proven to be increasingly useful in unexpected
ways. In the area of public health, Google Flu Trends, which analyzes routine Google queries in
order to provide a useful and timely supplement to conventional flu tracking methods, is a widely
publicized example of the unexpected uses of data.8 More generally, scientists are rapidly
developing methods capable of mining and modeling new data sources and “big” data. Thus
wider access to data and new forms of data collection are rapidly changing the study of humans,
human behavior, and human institutions, and these changes are sparking scientific progress.9
We argue that workplace injury and illness records should be made more widely available;
however, the records released to the public should not include information that is potentially
private and sensitive without additional safeguards. OSHA has a responsibility to apply best
practices to manage and mitigate potential individuals harms that might arise from its data
releases.
Scope of information made public
If released, the information that OSHA intends to make public under the Proposed Rule could
expose individuals to substantial re-identification risks. The legal standards cited by OSHA are
insufficient for gauging privacy risks and protecting employees against re-identification in
publicly-released records. The Proposed Rule may also prohibit the release of useful data
associated with relatively low re-identification risks. Thus, the approach fails to systematically
assess and balance re-identification risk, sensitivity, and utility of the data.
The Proposed Rule states that “OSHA intends to make public all of the collected data that neither
[the Freedom of Information Act (FOIA)], the Privacy Act, nor specific Part 1904 prohibit from
release.”10 According to OSHA, “all data fields in [Form 300A (summary log)],” “except for
Column B (the employee’s name), all fields . . . on [Form 300 (the log)],” and “all fields on the

7

Id.
Justin R. Ortiz, et al., “Monitoring Influenza Activity in the United States: A Comparison of Traditional
Surveillance Systems with Google Flu Trends,” PloS One 6.4 (2011), doi:10.1371/journal.pone.0018687; Nick
Wilson, et al., “Interpreting Google Flu Trends Data for Pandemic H1N1 Influenza: The New Zealand Experience,”
Eurosurveillance 14.44 (2009): 429-433; Samantha Cook, et al., “Assessing Google Flu Trends Performance in the
United States during the 2009 Influenza Virus A (H1N1) Pandemic,” PloS One 6.8 (2011),
doi:10.1371/journal.pone.0023610.
9
Micah Altman and Kenneth Rogerson, “Open Research Questions on Information and Technology in Global and
Domestic Politics – Beyond ‘E-,’” PS Political Science and Politics 41.4 (2008): 1-8; David Lazer, et al., “Life in
the Network: The Coming Age of Computational Social Science,” Science 323.5915 (2009): 721; Gary King,
“Ensuring the Data-rich Future of the Social Sciences,” Science 331.6018 (2011): 719-721.
10
Improve Tracking of Workplace Injuries and Illnesses, 78 Fed. Reg. 67254, 67263 (proposed November 8, 2013).
8

3

right-hand side of [Form 301 (incident report)] (items 10 through 18)” could be made publicly
available.11 This excludes, according to OSHA, personal information such as name, address, date
of birth, and gender from its publication requirements, but allows the publication of individuallevel information such as the employee’s job title, the date and time of the incident, and
descriptions of the injury or illness and where and how it occurred. Although neither FOIA nor
the Privacy Act likely prohibits the release of the information described above, it is not clear that
these laws are suitable benchmarks for determining the scope of information appropriate for
public disclosure.
The Privacy Act applies to “system[s] of records . . . from which information is retrieved by the
name of the individual or by some identifying number, symbol, or other identifying particular
assigned to the individual.”12 Since the database described in the Proposed Rule is based on
establishments, not individuals, any standards or conditions for information disclosure in the
Privacy Act are irrelevant since they are not readily applicable to the information in the Proposed
Rule. However, if the Privacy Act was deemed to apply to OSHA’s proposed database, it is not
clear that OSHA could release records to the public without written authorization from the
employees as the conditions of disclosure are limited.13
Similarly, FOIA seems to be of limited use as a standard to protect privacy in this instance.14
Under FOIA, U.S. agencies are required to release information in response to a request, unless
the agency determines that one of nine statutory exemptions applies.15 Determinations on
whether a particular exemption applies to information sought in a FOIA request are made on a
case-by-case basis. Although one FOIA exemption seems on point – it exempts the release of
information that “would constitute a clearly unwarranted violation of personal privacy”16 – FOIA
provides no guidance for determining the circumstances that might qualify as a “clearly
unwarranted invasion.” Agencies have wide discretion in such determinations. More troubling,
however, is that FOIA does not explicitly mandate that information subject to an exemption
cannot be released; rather, these exemptions are discretionary and may be waived by the agency
unless another law would prohibit release. In other words, OSHA appears to have substantial
deference in determining the scope of information that can be released in response to a FOIA
request. FOIA is also problematic as a standard because it is architected as a request-response

11

Id. at 67259-60.
5 U.S.C. § 552a(a)(5).
13
5 U.S.C. § 552a(b).
14
Legal scholars have also raised a number of concerns regarding the privacy protections for individuals in FOIA.
See, e.g., Evan M. Stone, “The Invasion of Privacy Act: The disclosure of my information in your government file,”
Widener Law Review, vol. 19, p. 345 (2013); Lisa Chinai, “Picture Imperfect: Mug shot disclosures and the
Freedom of Information Act,” Seton Hall Circuit Review, vol. 9, p. 135 (Spring 2013); Ira Bloom, “Freedom of
Information Laws in the Digital Age: The death knell of informational privacy,” Richmond Journal of Law &
Technology, vol. 12, p. 9 (Spring 2006).
15
5 U.S.C. § 552(b).
16
5 U.S.C. § 552(b)(6); State v. Washington Post Co., 456 U.S. 595 (1982).
12

4

system in which requests are individually reviewed – not for a system in which unstructured
information in free form text fields is categorically released to the public. The Proposed Rule
does not indicate whether similar review mechanisms would be available to weigh privacy risks
against the benefits of public disclosure before or after the release of information (e.g., fields 1018 in Form 301), which seems problematic since much of the information is descriptive yet
unstructured.
The other restrictions on release in Part 1904 also seem an insufficient standard for protecting
employees privacy interests under the Proposed Rule. Employers are required to omit employee
names and other descriptive information that might be used to re-identify an employee only in
certain types of incidents or illnesses. These are known as “privacy concern cases.”17 The scope
of this restriction is limited to the types of cases specified in the regulation that involve injuries
and illnesses of a sensitive character, such as “an injury or illness to an an intimate part of the
body or the reproductive system,” “sexual assault,” “mental illness,” “HIV infection, hepatitis, or
tuberculosis.”18 Outside of these special privacy cases, employers are not directed by the
regulations to limit the information in a way that would safeguard against re-identification,
which seems underinclusive for purposes of the Proposed Rule.
As discussed in the sections that follow, many examples in current literature suggest that the
combination of information that would be released under the Proposed Rule is likely uniquely
identifying for many employees in the data. In fact, some individual entries for a field, such as a
job title held by only one person at a company or a description of an unusual injury, may be
identifying on their own. The Proposed Rule’s standard for releasing or suppressing certain
categories of information is weak compared to other federal regulations,19 and there is no
indication that OSHA intends to provide additional safeguards for the information that is
publicly released. It also does not propose restrictions – technical, legal, or otherwise – on how

17

29 C.F.R. §§ 1904.29, 1904.35.
29 C.F.R. § 1904.29(b).
19
Although the proposed data disclosures are likely not governed by the Confidential Information Protection and
Statistical Efficiency Act (CIPSEA) or the Health Insurance Portability and Accountability Act (HIPAA), it is worth
noting that these laws rely on definitions of personally identifying information that are significantly more expansive
than the approach from the Proposed Rule. CIPSEA guidance states that “confidential information refers to any
identifiable information, regardless of whether direct identifiers such as name and/or address have been removed
from the individual records.” Office of Management and Budget, Implementation Guidance for Title V of the EGovernment Act, Confidential Information Protection and Statistical Efficiency Act of 2002 (CIPSEA) (Washington:
Office of Management and Budget, 2006), available at
http://www.whitehouse.gov/sites/default/files/omb/assets/omb/inforeg/proposed_cispea_guidance.pdf. In addition,
the HIPAA Privacy Rule states that individually identifiable health information is information that “relates to the
past, present, or future physical or mental health or condition of an individual; the provision of health care to an
individual; or the past, present, or future payment for the provision of health care to an individual” and that
“identifies the individual or with respect to which there is a reasonable basis to believe the information can be used
to identify the individual.” 45 C.F.R. § 160.103. Either of these standards would prohibit the release of information
such as a job title or an injury or illness description, to use just the examples above, that could reasonably be tied to
an individual.
18

5

the records containing these types of uniquely identifying information may be used by the
public.20
At the same time, the Proposed Rule calls for the routine redaction of information that, by itself,
is unlikely to pose a heightened re-identification risk. For example, checkbox items 8 and 9 on
Form 301, indicating whether an injury resulted in an overnight hospital stay or emergency room
visit, provide information about the severity of the injury. These fields are less likely to be
identifying than other fields that are proposed to be released. This redaction is substantively suboptimal since it reduces the utility of the released data for scientific and policy analysis. It is also
an indication that the Proposed Rule’s classification of fields as personally identifying or nonpersonally identifying is arbitrary.
Re-identification risks
The Proposed Rule’s requirement that certain fields be removed prior to disclosure is inadequate
as a de-identification standard based on the current literature on re-identification science. It is
now widely recognized that robust de-identification of microdata by simply removing and
generalizing fields is quite difficult. Guaranteeing privacy protections when releasing openended textual data, such as fields 14 to 17 on Form 301, is especially challenging. The risks
associated with de-identified microdata have led to a heated debate among privacy law scholars
about how to balance the risks, sensitivity, and utility of data when sharing it with third parties.21
There are many examples of datasets that were believed to have been de-identified but were later
re-identified. For instance, Latanya Sweeney recently demonstrated how news stories and public
records can be used to identify patients in hospital records that have been anonymized by
removing patient names and addresses.22 Using information such as name, age, residence, gender,
hospital, incident date, and details about the incident from news stories, as well as date of birth
and ZIP code from online public records, she was able to successfully identify a significant
number of individual hospitalization records in state hospital data. As another example, Kobbi
Nissim and Eran Tromer recovered records of over one thousand individuals in an anonymized
survey by querying an internet access mechanism hosted by the Israel Central Bureau of
Statistics web site and, furthermore, demonstrated that it is possible to link these records to

20

For example, a system that restricts access to the most sensitive data to only trusted users through technical means
coupled with legal contracts specifying additional conditions on use (e.g., re-sharing of data, publishing identifying
information, etc).
21
See, e.g., Paul Ohm, “Broken Promises of Privacy: Responding to the Surprising Failure of Anonymization,” 57
U.C.L.A. Law Review 57 (2010): 1701; Jane Yakowitz, “Tragedy of the Data Commons,” Harvard Journal Law &
Technology 25 (2011): 1; Ann Cavoukian and Khaled El Emam, “Dispelling the Myths Surrounding Deidentification: Anonymization Remains a Strong Tool for Protecting Privacy,” Information & Privacy Commissioner
(June 2011).
22
Latanya Sweeney, “Matching Known Patients to Health Records in Washington State Data” (July 2013),
available at http://dataprivacylab.org/projects/wa/1089-1.pdf.

6

individuals.23
The above examples from the re-identification literature illustrate that it is often possible to
identify individuals in a data set even after fields such as name, address, gender, and date of birth
have been removed. In fact, OSHA regulations also note that the descriptions of injuries and
illnesses may be identifiable on their own and encourage employers to exercise discretion in
describing the injury or illness in a privacy concern case if they “have a reasonable basis to
believe that information describing the privacy concern case may be personally identifiable even
though the employee’s name has been omitted.”24 However, the Proposed Rule would require
OSHA to release information that, in combination, would likely be uniquely identifying for
many of the individuals in the database. This means a friend, family member, colleague,
prospective employer, or even a marketer could potentially use personal knowledge of a
workplace incident or information from a news article to re-identify an individual in the OSHA
database and uncover sensitive details about the extent of her injury or illness.
Making workplace injury and illness records available while also providing stronger privacy
protections for employees will require a sophisticated approach to data sharing. Factors such as
re-identification risk, sensitivity, and utility of the data must be carefully considered and
balanced when designing a system for publicly sharing personal data.
Information sensitivity
Sensitivity of the information to be released is one of the factors that should be weighed against
the value and utility of the data in an information release. Generally, regulation should treat
information as sensitive when that information, if linked to a person, is likely to cause substantial
harm. There is a broad range of informational harms that are recognized by regulation and by
researchers and practitioners in the behavioral, medical, and social science fields.25 Types of
potential harms include loss of insurability, loss of employability, criminal liability,
psychological harm, social harm to a vulnerable group (e.g. stereotyping), loss of reputation,
emotional harm, and dignitary harm.
Although much of the information that would be released in accordance with the Proposed Rule
would likely be benign, there are some situations in which details regarding an injury or illness
may be sensitive. OSHA regulations already provide additional protection for “privacy concern
cases,” which include injuries or illnesses related to sexual assault, mental health, or infectious

23

Amitai Ziv, “Israel's ‘Anonymous’ Statistics Surveys Aren't So Anonymous,” Haaretz (Jan. 7, 2013),
http://www.haaretz.com/news/national/israel-s-anonymous-statistics-surveys-aren-t-so-anonymous-1.492256.
24
29 C.F.R. § 1904.29.
25
Elizabeth A. Bankert and Robert J. Andur, Institutional Review Board: Management and Function (Boston: Jones
and Bartlett, 2006); Raymond M. Lee, Doing Research on Sensitive Topics (London: SAGE, 1993).

7

diseases.26 However, OSHA’s exhaustive list of privacy concern cases does not include all
categories of sensitive information. There are additional types of cases that involve sensitive
issues, such as drug and alcohol abuse, and the public disclosure of such information would
create substantial privacy risks and potential harms for the individuals involved.
Review, reporting, and information accountability
The Proposed Rule also appears to lack review mechanisms, such as case-by-base redactions of
personal information that is particularly sensitive or vulnerable to re-identification risks. OSHA
would not have to look far to find examples of such review mechanisms. Current OSHA
regulations require employers to review and remove “personally identifying information” before
sharing workplace injury and illness records with non-governmental or contracted third parties.27
OSHA regulations also instruct employers to exercise discretion in describing the injuries and
illnesses in privacy concern cases.28
OSHA does not seem to require similar review mechanisms before the release of this information
under the Proposed Rule. The Proposed Rule calls for the public release of information such as
the time, date, and circumstances of injury, even though these descriptions alone may, in some
cases, be personally identifying. However, it does not indicate how, if at all, OSHA plans to
systematically review and redact personally identifying information from the descriptions in
these fields, or to prevent private information from being easily inferred by such redactions.
OSHA may not even be aware of the extent to which identifying information might be a problem
in the descriptive fields, given that it does not routinely access or collect Forms 300 and 301
outside of the limited number of investigations and inspections it conducts each year.
Moreover, the Proposed Rule appears to lack mechanisms that would provide accountability for
harm arising from misuse of disclosed data. Transparency, restrictions on disclosure, and
accountability for misuse are all essential to achieving an optimal balance of social benefit and
individual privacy protection.29 Accountability mechanisms should enable individuals to find out
where data describing them has been distributed and used, set forth penalties for misuse, and
provide harmed individuals with a right of action.

26

The privacy concern cases include “[a]n injury or illness to an intimate body part or the reproductive system; [a]n
injury or illness resulting from a sexual assault; [m]ental illnesses; HIV infection, hepatitis, or tuberculosis;
[n]eedlestick injuries and cuts from sharp objects that are contaminated with another person's blood or other
potentially infectious material . . . ; and [o]ther illnesses, if the employee voluntarily requests that his or her name
not be entered on the log.” 1904.29(b)(7).
27
Specifically, employers must “remove or hide the employees’ names and other personally identifying information”
before disclosing information on Forms 300 and 301 to third parties.” 29 C.F.R. § 1904.29(b)(10).
28
In certain “privacy concern cases” employers are encouraged to exercise discretion in “describing the injury or
illness” on both the OSHA 300 and 301 forms” if they “have a reasonable basis to believe that information
describing the privacy concern case may be personally identifiable even though the employee’s name has been
omitted.” 29 C.F.R. § 1904.29.
29
Daniel J. Weitzner, et al., “Information Accountability,” Communications of the ACM 51.6 (2008): 82-87.

8

Data sharing requires a nuanced approach
Addressing privacy risks requires a sophisticated approach, and the Proposed Rule does not take
advantage of advances in data privacy research or the nuances they provide in terms of dealing
with different kinds of data and finely matching sensitivity to risk. Like treatment of other risks
to subjects, treatment of privacy risks should be based on a scientifically informed analysis that
includes the likelihood of such risks being realized, the extent and type of the harms that would
result from realization of those risks, the efficacy of computational or statistical methods to
mitigate risks and monitor access, and the availability of legal remedies to those harmed.30
A modern approach to privacy protection recognizes the following three principles:
● The risks of informational harm are generally not a simple function of the presence or
absence of specific fields, attributes, or keywords in the released set of data. Instead,
much of the potential for individual harm stems from what one can learn about
individuals from the data release as a whole when linked with existing data.
● Redaction is often neither an adequate nor appropriate practice, and releasing less
information is not always a better approach to privacy. As explained above, simple
redaction of information that has been identified as sensitive is often not a guarantee of
privacy protection. In addition, the act of redacting certain fields of a record may reveal
the fact that a record contains sensitive information, and this fact is often sensitive by
itself or may lead to the deduction of sensitive information.31 Redaction may also reduce
the usefulness of the information, and alternative data-sharing methods, such as those
below, can often enable analyses that would be impossible with redacted data (while
providing stronger privacy protections).
● Thoughtful analysis with expert consultation is necessary in order to evaluate the
sensitivity of the data collected and their associated re-identification risks and to design
useful and safe release mechanisms. Naïve use of any data sharing model, including
those we describe below, is unlikely to provide adequate protection.
The examples in the next section describe how these three principles can be incorporated in an
information release that represents a modern, nuanced approach to privacy-protecting data
sharing.

30

Salil Vadhan, et al., Re: Advance Notice of Proposed Rulemaking: Human Subjects Research Protections (2010),
available at http://dataprivacylab.org/projects/irb/Vadhan.pdf.
31
See Krishnaram Kenthapadi, Nina Mishra, and Kobbi Nissim, “Denials Leak Information: Simulatable Auditing,”
Journal of Computer and System Sciences 79.8 (2013): 1322-1340.

9

Improved data sharing models
Below we provide a list of examples of data sharing models that can be used in conjunction with
new privacy-protecting methods to provide stronger privacy protections for subjects than deidentified microdata. The examples include:
● Contingency tables are tables giving the frequencies of co-occurring attributes. For
example, a 3-dimensional contingency table based on Census data for Norfolk County,
Massachusetts might have an entry listing how many people in the population are female,
under the age of 40, and rent their home.
● Synthetic data are “fake” data generated from a statistical model that has been developed
using the original data set. Methods for generating synthetic data were first developed for
filling in missing entries, and are now considered attractive for protecting privacy (as a
synthetic dataset does not directly refer to any “real” person).32
● Data visualizations are graphical depictions of a dataset’s features and/or statistical
properties. Data visualizations are especially useful for comprehending huge amounts of
data, perceiving emergent properties, identifying anomalies, understanding features at
different scales, and generating hypotheses.33
● Interactive mechanisms are systems that enable users to submit queries about a dataset
and receive corresponding results. The dataset is stored securely and the user is never
given direct access to it, but such systems can potentially allow for very sophisticated
queries. For example, the Census Bureau’s online Advanced Query System allows users
to create their own customized contingency tables.34
All of the models outlined above – privacy-aware methods for contingency tables, synthetic data,
data visualizations, interactive mechanisms – have been successfully used to share data while
protecting privacy, with no major compromises as far as we know. Each of these mechanisms
can be implemented with privacy protections and, when made privacy-aware in an appropriate
way, can provide strong protection. Conversely, each of these methods, used naïvely, can create
privacy risks. Many of these forms of data sharing have even been shown to be compatible with
a strong new privacy guarantee known as differential privacy.35

32

Donald B. Rubin, “Discussion of Statistical Disclosure Limitation,” Journal of Official Statistics 9 (1993), at 46168; Stephen E. Fienberg, “Conflicts Between the Needs for Access to Statistical Information and Demands for
Confidentiality,” Journal of Official Statistics 10 (1994): 115-32; John M. Abowd and Lars Vilhuber, “How
Protective Are Synthetic Data?,” Privacy in Statistical Databases (2008): 239-49.
33
Colin Ware, Information Visualization: Perception for Design (Boston: Morgan Kaufmann, 2004).
34
U.S. Census Bureau, Census Confidentiality and Privacy: 1790-2002 (2004), available at
http://www.census.gov/prod/2003pubs/conmono2.pdf.
35
Cynthia Dwork, “A Firm Foundation for Private Data Analysis,” Communications of the ACM (2011): 1, 86-95.

10

It is clear that we would want to make some of the above forms of sharing an option for
members of the public and for researchers when they would offer both better privacy and better
utility than de-identified microdata. Although no form of sharing is completely free of risk, tiered
access is required to ensure privacy and utility for different types of uses.36 It is likely that
publishing workplace injury and illness data using multiple levels of access would bring gains in
both privacy and utility.
Generally, for data that is made available to the public without significant restriction we
recommend that the data release process and method should ensure that no individual incur more
than a minimal risk of harm from the use of his or her data, even when those values are
combined with other data that may be reasonably available.
On this end of the privacy/utility spectrum, the unrestricted public release of data might be
limited to aggregate information. Such a release could be similar in detail to the aggregated
information currently provided by OSHA but cover all of the firms that would be required to
submit under the proposed changes. Further these kinds of aggregate statistics, can be released
with both formal guarantees of privacy and accuracy using existing differentially private
methods.37 The fact that large companies will likely have large numbers of incidents means that
the added noise won't harm the accuracy of the statistics by much.
We speculate that many member of the public would be likely to find that a series of contingency
tables and visualizations could simplify their review and comparison of the workplace safety
records of various employers. Within such aggregated releases, coding open-ended fields such as
injury and illness descriptions could additionally reduce the risk that sensitive details about an
individual’s injury or illness will be revealed.
An intermediate level of access could be set up to enable interactive analysis of the data, through
a privacy-aware model server. This server would ensure that the results provided by the analysis
leak minimal private information. It could also be used to permit audits of access and to impose
some click-through data use agreements providing individuals with additional legal protections
from misuse.
At the same time, for a user to gain the full utility of the data, she must have rich access to
information that is minimally redacted and at the finest level of granularity obtainable. In cases
where such access is needed, it should be provided through a protected and monitored data
environment, such as a virtual (remote-access) data enclave,38 and complemented with legal

36

See National Research Council reports, supra note 4.
Cynthia Dwork, Frank McSherry, Kobbi Nissim, Adam Smith: Calibrating Noise to Sensitivity in Private Data
Analysis. TCC 2006: 265-284
38
Julia Lane and Stephanie Shipp, “Using a Remote Access Data Enclave for Data Dissemination,” International
Journal of Digital Curation 2.1 (2008): 128-134.
37

11

agreements providing information accountability and appropriate restrictions on use and sharing
of the data.
Finally, OSHA should consider developing a toolkit or educational materials to help employers
identify information that poses a re-identification risk in their workplace records, especially if
OSHA expect that its recordkeeping forms will continue to elicit textual descriptions of injuries
and illnesses in the future. Such materials could help mitigate the risk that employers will include
identifying information in the forms.
Conclusion
We argue that workplace injury and illness records should be made more widely available
because releasing these data has substantial potential individual, research, policy, and economic
benefits. However, OSHA has a responsibility to apply best practices to manage data privacy and
mitigate potential harms to individuals that might arise from data release.
The complexity, detail, richness, and emerging uses for data create significant uncertainties
about the ability of traditional ‘anonymization’ and redaction methods and standards alone to
protect the confidentiality of individuals. Generally, one size does not fit all, and tiered modes of
access – including public access to privacy-protected data and vetted access to the full data
collected – should be provided.
Such access requires thoughtful analysis with expert consultation to evaluate the sensitivity of
the data collected and risks of re-identification and to design useful and safe release mechanisms.

12

Privacy tool osha comments

More Related Content

What's hot (20)

Similar to Privacy tool osha comments (20)

More from Micah Altman (20)

Recently uploaded (20)

Privacy tool osha comments