Table 3.

Studies of IMIS exposure data having reported results related to potential biases.

Publication	Main focus	Exposure metric	Variables studied	Bias
Oudiz et al. 1983	silica exposures in foundries	% of exposures above PEL, severity	work area, type of foundry, number of employees	fraction of overexposures increasing with number of employees
Jones 1986	under reporting in IMIS	% of samples in OSHA reports ending up in IMIS	N.A.	slightly fewer than 50% of compliance data reported in IMIS, 25% of plants with compliance data do not appear in IMIS, under-reporting does not seem related to level of exposure
Froines et al. 1986a	general portrait of silica exposure	severity	industry, union status, inspection type, job description	despite between-industry differences, general trend of higher probability of being >PEL for complaint inspections, especially in unionized companies. No consistent trend for mean severity.
Stewart et al. 1990	use of IMIS for occupational epidemiologic studies	concentration	industry, job description	SIC specific measurement arithmetic mean higher for complaint inspections (median ratio of 2.4, 3 out of ten ratios less than 1)
Froines et al. 1990	general portrait of lead exposure	median severity by inspection	industry, number of employees, union status, inspection type	odds ratio of 3 for complaint inspections versus scheduled for the probability of a median severity within an inspection to be greater than 1
Gomez 1997	association between IMIS variables and reported exposure levels	concentration, company specific mean concentration, probability of being greater than a specified value	job description, number of employees, union status, year, scope of inspection, type of inspection	clear trend for number of employees (exposure level decrease when number of employees increase : GMs for large companies (>273 employees) are 30–40% of those from small company (<60 employees)
Tanner-Martinez 1997	effect of non-random sampling on estimation of exposure variability from IMIS data	company-specific geometric standard deviation	auto-correlation structures	GSDs smaller when estimated from few samples (n smaller than 6) or from samples within a small time period (week)
Melville and Lippman 2001	association between IMIS variables and reported exposure levels	concentration, company specific mean concentration, probability of being greater than a specified value	job description, number of employees, union status, year, scope of inspection, type of inspection	variable results. General trend of higher levels for general scope inspections. For toluene and formaldehyde, levels associated with complaint inspections higher versus scheduled. Quantitative estimates no provided.
Lurie and Wolfe 2002	general portrait of exposure to hexavalent chromium	concentration, number of measurements, citations	year, industry, inspection type, inspection conducted by federal or state agency	greater % of non-detects in state inspections (59.8% versus 48.9%) compared to federal inspections.
Middendorf 2004	surveillance of occupational noise exposure	several noise exposure metrics	year, number of employees,	noise levels increase with number of employees (shift of 2–3 dBA from <20 to >499 employees). Mean consultation levels > mean enforcement levels (up to 4 dBA depending on year, average ~2)
Okun et al. 2004	trends in occupational lead exposure	probability of a measurement exceeding the PEL	year, region, number of employees, union status, inspection type	probability of being higher than PEL slightly higher for compliance data than for consultation data (between 1 and 5% across years), and for complaint inspection than for general schedule inspections (estimate of 5% from logistic regression)
Yassin et al. 2005	general portrait of exposure to silica dust	concentration	year, industry, job description, inspection type	programmed inspection industry specific geometric means slightly higher than overall industry specific GMs (0.077 versus 0.073mg/m3)
Lavoué et al. 2008	general portrait of exposure to formaldehyde	concentration	inspection type, sample type (short-term, TWA), season, industry, year, state, outside temperature,	marginal effect of inspection type with complaint and referral inspections associated with slightly higher levels than scheduled inspections (7%). Exclusion of non-detects might have caused underestimation of ~20–30% for TWA results, up to 60% for short-term results.
Lavoue et al. 2011	comparison of formaldehyde exposure levels in IMIS and the French exposure databank COLCHIC	concentration	data source, year, sample type (short-term versus TWA), industry	formaldehyde levels somewhat higher in the French database (by 14% in average, reduced to no difference after exclusion of health sector). Contrast between most industries very similar. Exclusion of non-detects would have caused overestimation of IMIS TWA results by ~20% and underestimation of the COLCHIC short-term data by ~30%.
Henn et al. 2011	general portrait of exposure to lead	percent of samples over the PEL	industry, time period, region, number of employees, federal/state plan, union status, inspection type, advance notice of inspection, presence of employee representative, employees interviewed during inspection	higher probability of being over the PEL for smaller companies (1–99 versus over 500 :OR=2), federal versus state plan (OR=1.1), union versus no union (OR=1.23), advance notice of inspection (OR=1.6), absence of employee representative(OR=1.19), no employee interviewed (OR=1.33)
Teschke et al. 1999	exposure to wood dust for a population-based case–control study	concentration	year, job description, number of employees, inspection type	none reported in multivariate analysis. In univariate analysis, GM for planned inspection slightly lower than program related (complaint or referral, 1.86 versus 1.99mg/m3)