2. Background

While often conflated, drilling and HF are separate processes: drilling refers to the entire process of developing a hydrocarbon resource, while HF is a specific part of the process. Fracturing is performed with specialized pressure pumping equipment, usually after the drilling rig is removed. The wellbore is perforated in a target formation. To focus available pumping power on isolated sections of the wellbore, the pressure pumping typically occurs in stages, with each stage focusing on a portion of the perforated wellbore. A slurry of base fluid (usually water), proppant (often sand), and chemical additives is mixed on the surface. The pressure pumps then force the slurry into the target area, creating complex fractures in the hydrocarbon-bearing substrata. As pressure is reduced, proppant remains in the fractures, leaving a pathway for oil and gas to migrate to the wellbore and then to the surface. Oil and gas molecules that would otherwise be trapped underground can thereby be brought to the surface.Footnote ³

Each frac job has many variables: the amounts of water and sand, the maximum pressures and duration of pumping, and the combination of additives to the injected fluid, which may have a particular recipe. Additives are intended to enhance performance of the fluid in one or more dimensions, for example by reducing friction or increasing viscosity.Footnote ⁴ The recipes for injected ingredients are viewed as important proprietary assets, especially by service companies. Service companies advertise various recipes and consulting expertise to operating firms.

A substantial technical literature addresses aspects of fracture design and implementation. We focus on two stylized facts: first, that fracture designs continue to evolve. Patel et al. (Reference Patel, Robart, Ruegamer and Yang2014) compares the time trends in frac fluid and proppant design in the Marcellus, Bakken, and Denver-Julesburg basins in the US. In each of these locations, firms have continued to experiment with alternative proppants and worked to increase the amount of proppant delivered. Second, that alternative fluid configurations are used in different areas. This makes cross-sectional analysis of fracturing fluid difficult. What works in the Marcellus may be totally inappropriate in the Bakken. This leads us to focus on a single geographic area, where we use relatively small variation in fracturing fluids and designs to identify differences in production. More importantly, both of these stylized facts have important implications for the propagation of HF and unconventional natural gas development more generally.Footnote ⁵

Fetter et al. (Reference Fetter, Steck, Timmins and Wrenn2018) uses regulatory disclosure data from Pennsylvania to document the diffusion of the most productive frac recipes to other firms, and an attendant reduction in new experimentation that was previously led by the most productive firms. This poses a tradeoff for regulators: requiring disclosure of chemical additives to assuage environmental concerns allows other firms to imitate the most productive firms in the sector, but also curtails innovation by those same firms. A regulatory scheme that mandates disclosure or compromises the productivity gains that firms capture by experimenting with frac recipes could undermine the viability of unconventional resources. On the other hand, excessive secrecy may exacerbate fears about environmental degradation even as productivity gains could be realized.

The widespread endowment of natural gas deposits that could be exploited with HF creates an opportunity for countries across the developing world. Natural gas resources, particularly unconventional resources that will likely require fracturing, are abundant throughout the developing world (Weijermars et al., Reference Weijermars, Drijkoningen, Heimovaara, Rudolph, Weltje and Wolf2011; Bocora, Reference Bocora2012; Wang et al., Reference Wang, Ma, Tong, Liu, Zhang, Wu, Li, Wang, Xie and Yang2016; Evensen, Reference Evensen2018). However, it is important to recognize that continual improvement and modification are commonplace, as the design and application of frac jobs is adjusted to best suit the particular geology. Specific countries where HF is thought to have the potential to greatly increase production include Argentina (Mares Reference Mares2014a, Reference Mares2014b, Reference Mares2014c, Gomes and Brandt, Reference Gomes and Brandt2016; Sierra, Reference Sierra2016; Florencia, Reference Florencia and Gamper-Rabindran2018), Brazil (Dora, Reference Dora2014), China (Lin, Reference Lin and Gamper-Rabindran2018), Mexico (Mares, Reference Mares2012), and South Africa (Morkel and de Witt, Reference Morkel, de Witt and Gamper-Rabindran2018).

Argentina is a particularly noteworthy example, as it is believed to possess the second-largest shale gas deposits and fourth-largest shale oil deposits in the world (Mares, Reference Mares2014b; Sierra, Reference Sierra2016). Interest in Argentina's shale resources has been strong following a significant shale resource discovery in 2011 by Yacimientos Petrolíferos Fiscales (YPF), the national oil company. Much of this interest has been expressed by oil and gas majors like Chevron, ExxonMobil, Shell, and Total (Mares Reference Mares2014a, Reference Mares2014b; Costie et al., Reference Costie, Holm and Berardo2018). Particularly important is the Vaca Muerta deposit, which has been referred to as the ‘crown jewel’ of Argentina's shale resources (Gomes and Brandt, Reference Gomes and Brandt2016). In 2013, the Argentine government revised its oil and gas regulations to attract investment (Mares, Reference Mares2014c) in order to promote development of its shale resources (as well as some conventional deposits). Development of Vaca Muerta began in earnest not long after, starting with an accord signed between Chevron and YPF, and ratified by the local government (Costie et al., Reference Costie, Holm and Berardo2018). Despite these reforms, chronic national policy uncertainty works against development of Argentina's resources (Mares, Reference Mares2013, Reference Mares2014b, Reference Mares2014c; Gomes and Brandt, Reference Gomes and Brandt2016).

Development of the Vaca Muerta shale deposit seems to present an intriguing example of the tension between the potential benefits from expanded production and the possible costs of development, which are thought likely to fall on citizens near the zone of development. Environmental concerns have also been expressed. Mares (Reference Mares2012) notes a ‘growing NGO movement in the country that is focusing on HF issues.’ Similarly, Sierra (Reference Sierra2016) opines that environmental groups and local communities are likely to oppose HF until the government establishes a suitable regulatory framework for unconventional production; Gomes and Brandt (Reference Gomes and Brandt2016) makes a similar point. Relatedly, Costie et al. (Reference Costie, Holm and Berardo2018) argue that ‘antifracking coalition members responded to the [Chevron-YPF] accord by becoming considerably more active.’ One argument for the strength of such opposition is the belief that understanding of the risks associated with HF is weak. Indeed, Mares (Reference Mares2013) argues that such ‘environmental concerns will slow shale gas development in Argentina.’ A key goal of our analysis is to offer greater clarity of some of the key aspects of unconventional resource development.

Mexico provides another example of potential gains from broader adoption of HF. Decline has overtaken replacement of Mexican oil and gas production (BP, 2019), even as pipeline imports of natural gas from the US increase. Some of this gas is produced from the Eagle Ford shale in South Texas, a formation that is thought to extend south across the border, but which has not yet been commercially exploited. Following recent energy sector reforms, the Burgos Basin, which encompasses the Eagle Ford resources, has been opened to private investment in a tacit admission that the national oil company Petròleos Mexicano (PEMEX) has not been able to easily tap those resources on its own.

In addition to the technical considerations associated with grafting the HF technology onto a new geologic setting, there are important concerns regarding the effects of HF on other resources, in particular fresh water. Our study is well positioned to clarify this debate, as we consider technical (productivity) related concerns, institutional (industrial organization) related concerns, and environmental (toxicity) related concerns.

3. Conceptual framework

A main focus of our analysis is to determine the role played by various ingredients in an HF process. To that end, we posit a production function that can be described by

(1)\begin{equation} \ln (y) = \beta \ln(A) + \gamma \ln (z) + \sum_{i} \alpha_i \ln (X_i), \end{equation}

where $y$ is the productivity measure of interest, $z$ is the footage of the wellbore that is treated by HF, $X_i, i = 1, \ldots, n$, is an ingredient used in the frac job (measured as volume of ingredient per foot treated), and $A$ is a well-specific productivity parameter. We assume $\ln (A)$ is a linear combination of the elapsed time since the well was drilled, $\textrm {AGE}$, the total volume injected in the well per foot treated, $SIZE$, and the ratio of proppant (typically sand) to water injected, $SANDRATIO$ (Covert, Reference Covert2014).

The data we use lists scores of additives; to facilitate the empirical analysis we aggregate into six categories: biocides, breakers and crosslinkers, chemical additives, gelling and foaming agents, surfactants, and all other ingredients. Table A1 in the online appendix summarizes the six classes of additives with examples.

As operators compete in competitive output markets, they have a choice of differentiated completion inputs provided by different service companies. One argument for product differentiation among service companies is to relax pressures to compete on price; to the extent that service companies disguise their respective product mixes, they are able to enhance market power.Footnote ⁶ Anecdotally, service companies are able to price discriminate, which implies that product differentiation should be greatest to support this outcome. Service companies could differentiate their products in a number of ways, but adjusting the frac recipe is one obvious mechanism.

4. Data

Our analysis relies on data drawn from four sources. First, we use well-specific records extracted from the FracFocus repository to document what was injected into each wellbore during the well completion process. FracFocus is a joint venture of the Interstate Oil and Gas Compact Commission and Groundwater Protection Council (see their site at http://www.fracfocus.org). Since 2010, FracFocus has published company-provided data on HF jobs across the country. In some states, FracFocus has satisfied state-level disclosure regulations; in Wyoming, the Wyoming Oil and Gas Conservation Commission (WOGCC) – the primary regulator of oil and gas operations in the state – maintains a similar but separate registry for legal disclosure purposes that is not available to the public. In Wyoming, all wells with drilling permits issued after August 17, 2010 were subject to disclosure.Footnote ⁷ The FracFocus database has been revised through three versions, and we use data from each version because wells were drilled and reported at varying times as the database evolved.Footnote ⁸ The information disclosed in each version of the FracFocus database has changed slightly, so we were careful to extract similar information from all versions.

Second, we used completion reports (Form 3) from the WOGCC to augment the FracFocus data with additional information about the frac jobs. The completion reports allowed us to cross-validate the FracFocus data on injected substances, as well as add information about technical aspects of each frac job. Operators are required to file Form 3 with the WOGCC after a well is completed and each time a well is recompleted. We dropped all wells with recompletions from our production analysis, but kept them for toxicity analysis.Footnote ⁹ Appended to these reports is detailed information about frac design and implementation. We use these data to characterize technical aspects of the completion, including the number of stages and the footage of the treated interval.

Third, we matched the specific injectant data from FracFocus to information about toxicity of the substances using the Occupational Safety and Health Administration (OSHA) Occupational Safety Database. This allows us to comment on the potential for environmental spillovers. The OSHA Occupational Chemical Database (available at https://www.osha.gov/chemicaldata/) includes 751 chemical compounds that are considered potentially hazardous. The database was originally compiled in cooperation with the Environmental Protection Agency. Compounds are indexed by unique chemical abstract service (CAS) numbers. Because these same unique identifiers are reported in FracFocus records, we are able to match the compounds directly. The database also reports exposure guidelines. In cases where we were unsure about a compound, we referred to the National Institute for Occupational Safety and Health pocket guide to clarify. Despite this cross-referencing, there were numerous compounds that we were not able to identify by reported name or CAS number. Our maintained assumption is that any such compounds are not toxic hazards.Footnote ¹⁰

Fourth, we used production reported at the well-month level to assess the ultimate production of alternative approaches to HF. These data were compiled by private data provider Enverus (www.enverus.com). These data are derived from operator reports of well-level production. Production can be aggregated in several different ways. In our preliminary results, we use aggregate production over the first 6 or 12 months of a well's life. We acknowledge that some elements of frac design may be targeted at increasing peak flow, or maintaining higher flow rates over time. We also used monthly data to estimate a panel model of well performance.

One of the most important production basins in Wyoming lies in Sublette County (figure A1 in the online appendix). Oil and gas have been part of the landscape there since 1939, when the first well was drilled there (though the well was plugged and abandoned after it found natural gas, as opposed to oil). By the early 1990s, advances in exploitation of western tight sands made the geology of Sublette County more attractive. HF of vertical wells was common at this point in time, although the designs were far different and smaller in scale than more recent application of the technology that we investigate. Production of both oil and natural gas expanded rapidly during the first years of the 2000s.Footnote ¹¹

An overwhelming majority of the wells in Sublette County are located on federal surface and minerals, and are therefore subject to federal environmental oversight as administered by the Bureau of Land Management (BLM). A very small fraction of wells are located on private minerals, with the balance on state-owned surface and minerals. All these wells are subject to environmental regulation by the state of Wyoming. During the period we study, federal agencies had no rules pertaining directly to HF, so state rules covered both.

We have detailed completion and production information from 569 wells in Sublette County, of which 297 wells have information from both FracFocus and WOGCC completion reports and have not been fractured more than once. Our sample is primarily drawn from two adjacent western tight sands plays that share a common leasing and well-servicing market.

As operators hire service companies to fracture, both parties may experience gains from learning-by-doing. Contemporaneously, firms may observe other experiments and tweak their own HF formulae to increase production. A listing of variables used in our analysis, along with their units of measurement, is contained in table A2, while table A3 provides summary statistics for frac jobs in our sample (see the online appendix). A wide range of frac designs are used, with an order of magnitude separating the smallest from the largest completions in our sample as measured by injected water volume. Similarly, an order of magnitude separates the lightest fracs (lowest sand/water ratio) from the highest. Suspending a greater quantity of proppant in a fixed amount of fluid requires increasingly sophisticated fluid chemistry.

An important question we address is: how common is it for additives to be withheld as confidential business information, or trade secrets? Table 1 summarizes the use of trade secret provisions across the categories of fluid additives. To simplify the complexity of frac fluids with dozens of ingredients, we use categories of additives characterized by purpose in Stringfellow et al. (Reference Stringfellow, Domen, Camarillo, Sandelin and Borglin2014). Many of the ingredients that we see in injection reports do not match with compounds analyzed by Stringfellow et al. (Reference Stringfellow, Domen, Camarillo, Sandelin and Borglin2014), and we therefore have a difficult time categorizing them. For compounds that we observed in other wells, we used the modal purpose in the other data to help classify ingredients by purpose. Nonetheless, we have many unclassified additives. The unclassified category is the most common category for trade secrets, followed by biocides used to control subsurface bacteria.Footnote ¹²

Table 1.

Count of ingredients per frac job withheld as trade secrets

Notes: Data compiled from FracFocus records. $N=297$. Total includes undisclosed proppants.

Well completions occur over a period of four calendar years (2010–2013), which raises the possibility that a trend in claims of secret additives occurred. While we did not find a significant trend over time, we do observe significant differences across firms – both operators and service companies. Disclosure of the fluid ingredients reveals to the public and potential competitors what is in fracturing fluid. The FracFocus records provide information about the concentration of each ingredient in the additive and in the final fluid that is injected. This makes it hard to back out exactly how to mix the fluid and probably contains no information about sequencing or timing of mixing, but provides enough information to expect what concentration of a given additive is likely to be in the fluid (if, for example, an interested person wanted to know what calibration was needed to detect the presence of a particular chemical compound). While ingredients are important, the precise measurement and the practice of how they are combined is at least as important as knowing the identity of each component. The recalcitrance of some firms to reveal the ingredients has exaggerated concerns about the environmental risks associated with those additives.

There is a potential tradeoff between the trade secrecy and disclosure of a CAS number and associated toxicity. Recall that undisclosed additives cannot be assessed for toxicity using our methodology. A firm might have a number of motivations for invoking a trade secret. One reason is that a firm might believe it can maintain an edge in the marketplace by obscuring its particular recipe for HF. Alternatively, a firm might simply be experimenting with a new formula, and wishes to conceal the experiment and its outcome from any potential competitors. Third, and non-exclusively, a particularly toxic additive could be kept out of public scrutiny by protecting it with a confidential label. Weighing the various arguments for and against the disclosure of a particular additive is likely to vary at the firm level, as competitive position and views on corporate liability might vary.

Using our measure of toxicity, table 2 summarizes the number of different additives, by category, that are indicated to have toxicity hazards. As expected from previous studies, two of the most toxic categories of additives are gels and breakers or crosslinkers, which are used to adjust fluid viscosity (upwards for injection, then downwards for fluid recovery). More of a surprise is that proppants are more often matched to toxic substances than many other compounds. Certainly sand and ceramic proppants are not often thought to be the greatest environmental hazard in frac fluid. However, many operators do treat the proppant, such as coating it with resin; and the coating may be in the OSHA database. Chemical additives used to balance pH or other properties of the fluid, because they are often strong acids, are unsurprisingly also common on the toxic list.

Table 2.

Count of toxic ingredients per frac job

Notes: Data compiled from FracFocus records, as matched to OSHA Occupational Safety Database. $N=297$.

The wells in our sample produce some oil along with the gas, as well as valuable natural gas liquids (figure A2 in the online appendix). For that reason we consider both raw (dehydrated) gas production to identify the main target and total production on a barrel of oil equivalent basis to identify the whole slate of produced compounds. Products are generally not priced on an energy equivalent basis, but the conversion to BOE is useful to consider alternative measures of production.Footnote ¹³ Here, again, we find large differences in production between the best and worst wells. The connection between production differences and measurable differences in frac inputs is at the core of our productivity analysis.

Because we do not observe sales prices, it is not clear how long an average well must produce before drilling costs are recovered. As mentioned above, large basis differentials for western Wyoming are likely to affect the economic value of production. Firms will balance the the economic value of incremental production from adjusting the frac design or fluid with the cost of those changes. Our analysis falls short of being able to assess the profitability of such input adjustments, both because we do not observe the input costs, but also because we do not observe realized revenues.

The wells in our sample were operated by four firms at the time of completion, one of which has only a handful of the wells that we observe fully. The well completions were provided by three different service companies. In some cases subcontractors assist the primary service company or provide some services that we can observe. For the purposes of our analysis, we focus on the three major service companies, which are likely to have a more substantial role in the overall design of the completion. Table A5 (online appendix) summarizes the frequency of pairings between the operators that we observe and the service companies.

In our estimation sample, service companies A and B have roughly equal positions in the local market, each with about 20 per cent of the wells. Company C has a larger stake of about 45 per cent. There is a similar concentration among the operating firms – with three substantial firms. The two smaller of these three firms have roughly equal shares, while the third firm is much larger. The table also reveals specific relationships between operators and service companies. Operators II, III, and IV exclusively use one service company. Operator I uses multiple service companies, and also has wells that do not report a service company (and may have been vertically integrated).

Table 3 explores the relation between firm identities, withheld additives – ‘trade secrets’ – and toxicity of additives. We find significant differences in how different firms employ both trade secret provisions and toxic additives. While there are large differences across service companies and across operators, it is not clear from table 3 whether operator or service company is making the decision not to disclose particular additives. There is a motive for the service company (the seller in this relation) to obscure the mix used in the frac job, so as to enhance its market power (Carlton and Perloff, Reference Carlton and Perloff2008); this observation suggests it is the service company, and not the operator, which is electing to withhold some ingredients. On the other hand, ultimate responsibility for the well rests with the operator, including liability for environmental consequences, so it may be in their interests to create an air of obfuscation. Under these circumstances it might also be beneficial for the operator to allow the public to think that service companies are keenly interested in protecting trade secrets.

Table 3.

Summary of trade secrets and toxicity, per well by service company and operator

Notes: Data compiled from FracFocus and WOGCC, as matched to OSHA Occupational Safety Database. Total $N=297$.

5. Empirical strategy

We empirically investigate three dimensions of the fracking process in Sublette County. First, we use the detailed inputs used on different wells to understand the productivity impacts of adjusting the production process. Second, we assess the productivity effects of withheld additives to determine whether such additives can yield higher production. While these results are interesting and policy-relevant, our empirical study does not allow us to assess the net benefits of permitting additives to be withheld from public scrutiny. Third, we use our measures of toxicity of various additives to perform a similar analysis of how more toxic recipes are correlated with production. Like the secrecy analysis, these results do not provide the net benefits from permitting toxic additives to be injected, but do document the gains to be had from including some additives that can be toxic. Throughout each of these analyses, we are interested in the relationship between operating firms and their service company partners. The roles of these partners in enhancing productivity, determining which additives are used, and which are not disclosed helps shed light on how the industry has deployed technology to unlock the potential of unconventional resources.

5.1. Production outcomes

Our initial empirical specification is a cross-sectional analysis of aggregate production measure during the initial 6 or 12 months of a well's life, based on equation (1):

(2)\begin{equation} \ln y_{j} =\beta_0 + \beta_1 \textrm{WATER}_j + \beta_2 \textrm{SANDRATIO}_j + \sum_{i} \alpha_i \ln X_i + \sum_{k} \gamma_k D_{kj} + \varepsilon_{j}, \end{equation}

where $D_{kj}$ is an indicator variable equaling 1 if well $j$ was completed in year $2010+k,$ for $k =1,2,3$. We use the log of the volume of total base fluid injected as the basis for $WATER_j$. For every well we also know the location, service company, and the operator at the time of the completion. We also include the relative importance of different classes of additives as described in tables 1 and 2.

Because we have data on monthly production at each well, we also estimate a well-month panel specification,

(3)\begin{equation} \ln y_{jt} = \gamma \textrm{AGE}_{jt} + \beta' A_j + \sum_{i} \alpha_i \ln X_i + \varepsilon_{jt}. \end{equation}

The variable $\textrm {AGE}_{jt}$ measures the number of months well $j$ has been actively producing as of time $t$. Accordingly, the parameter $\gamma$ can be interpreted as the average monthly decline rate in production. To the extent that characteristics of the frac job affect the decline rate, interacting those measures with $\textrm {AGE}_{it}$ will provide inference about the dynamic effects.

5.2. Trade secrets and undisclosed additives

Beyond the basic specifications relating frac designs to production, we examine the impact of undisclosed additives. These regressions identify broad trends in the employment of trade secret provisions to avoid disclosure. We are especially interested in how different operators and service companies might differentially employ these tactics. We begin by estimating the production effects of withheld additives directly.

(4)\begin{equation} \ln y_{jt} = \beta' A_j + \sum_{i} \alpha_i \ln X_i \times \mathrm{SECRET}_i + \varepsilon_{j}. \end{equation}

The interpretation of the $\alpha _i$ in this specification is as in weighted least squares, with categorical withheld additive counts as weights. We augment those results with an analysis of the total incidence of withheld additives.

(5)\begin{equation} \sum \mathrm{SECRET}_{i} = f \left (A_i, \mathrm{OPERATOR}_i, \mathrm{SERVICE}_i, \mathrm{YEAR}_i \right). \end{equation}

Because the dependent variable is a non-negative integer, a count model is appropriate for this analysis; we use a negative binomial model.Footnote ¹⁴

5.3. Toxicity

The third dimension of frac jobs that we investigate is the role of toxic additives. Because it is hard to measure toxicity directly, we interact the aggregate toxicity in each additive class with the amount.

(6)\begin{equation} y_{jt} = \beta' A_j + \sum_{i} \alpha_i \ln X_i \times \mathrm{TOXIC}_i + \varepsilon_{jt}. \end{equation}

The interpretation of the $\alpha _i$ in this specification is as weighted least squares, with the categorical toxic additive counts as weights.

We also perform a count analysis of toxic additives similar to the count analysis for trade secrets.

(7)\begin{equation} \sum \mathrm{TOXIC}_{i} = f \left (A_i, \mathrm{OPERATOR}_i, \mathrm{SERVICE}_i, \mathrm{YEAR}_i \right). \end{equation}

Note that because the toxicity of withheld additives cannot be assessed, we are not able to consider interactions between secrecy and toxicity.

6. Empirical results

We now discuss our empirical analysis. We start with a brief discussion of the basic specification, equation (2), focusing on regression results based on the amount of natural gas produced in the first 6 or 12 months of production per thousand feet of treated interval. We find evidence of significant increases in production over time, which can be taken as evidence of learning about the technology – although these gains are not monotonic over time: there appears to be a peak in 2012; this could be attributable to depletion of high-production well locations within the county. We also find evidence of mean differences between operators and service companies, as well as spatial variation in production throughout the county. These results are broadly consistent with prior work.Footnote ¹⁵

Because natural gas production extends over a period of time, adjusting the frac job to improve production over time may provide greater returns than other efforts at product differentiation. Table 4 presents results of panel regressions along the lines of equation (3), where we also control for the number of stages and the ratio of sand to water and the total volume of water.Footnote ¹⁶ We find evidence of steep decline rates – 7.5 per cent per month – over the short period of observed production. This is consistent with previously estimated decline rates in Wyoming (Mason and Roberts, Reference Mason and Roberts2018). There is not much evidence that decline rates are directly affected by fluid additives as we measure them. Nor do toxicity or trade secrets have a lot of purchase in these specifications.

Table 4.

Panel results, log monthly gas production

Notes: Dependent variable is the log of monthly gas production. Operator clustered standard errors in parentheses. Significance: ***p < 0.01, **p < 0.05, *p < 0.1.

We next investigate whether withheld or toxic ingredients exhibit important productivity effects; results are reported in table 5. We focus here on the first six months of production; alternative measures are evaluated in tables A9 and A10 in the online appendix. We allow for idiosyncratic effects from each category of ingredients, along with interaction effects associated with withholding. We include indicator variables for operators I and II, along with the three different service companies; we also include indicator variables for the year the well was completed for 2011–2013 (so 2010 is the excluded year). Accordingly, the estimate associated with a year shows the difference in the average estimated number of withheld ingredients between that year and 2010 – an interesting point of comparison, as Wyoming's reporting regulation took effect in 2010. The results do not suggest strong productivity effects associated with either withholding or toxicity: the estimated cofficients are modest and generally statistically insignificant, both for ingredients that are withheld as trade secrets, and ingredients that are toxic.

Table 5.

Trade secret and toxicity results

Notes: Dependent variable: natural log of first 6 months’ gas production. Number of observations = 298. Robust standard errors in parentheses. Lower panel reports coefficients for interaction terms between additive type and secrecy (regression 2) or toxicity (regression 3). Significance: ***p < 0.01, **p < 0.05, *p < 0.1.

Our final line of inquiry relates to the number of ingredients used in a frac job. As noted above, we use a negative binomial count model.Footnote ¹⁷ Table 6 reports results of two different negative binomial models that address how many different additives are included in frac recipes. Column (1) reports the number of additives withheld as trade secrets as the dependent variable, while column (2) uses the count of the number of toxic additives as a dependent variable. These results point to some broad conclusions.

Table 6.

Negative binomial model results: secrecy & toxicity

Notes: Dependent variable in column (1) is count of all injected additives that are withheld as trade secrets. Dependent variable in column (2) is count of all injected additives that are identified as toxic by OSHA. Point estimates with robust standard errors reported in parentheses. Significance: ***p < 0.01, **p < 0.05, *p < 0.1.

First, we note that frac jobs that use more water tend to include fewer additives that are toxic, suggesting that there may be gross substitution between these inputs. This correlation hints at environmental implications: because smaller volumes of water are correlated with more toxic additives, these toxic additives are likely to be relatively more concentrated in the event they come into contact with the broader environment. This result takes on added significance in light of the multiple demands on water in the basin: Sublette County is an agricultural region that contains the headwaters of the Green River, a major tributary to the Colorado River, which is an important river system for the western United States. Substituting the demand for water may be offset with potentially more toxic frac recipes. Similar concerns may arise for water-stressed basins in developing countries; this seems likely to apply, for example, in the Vaca Muerta basin of Argentina.

Second, wells using a higher sand/water ratio tend to contain more toxic ingredients. A key use of the various ingredients in a frac job is to facilitate the transit of the proppant into fissures in the host rock; evidently, the companies in our sample preferred relatively more toxic ingredients for this task.

Third, increases in the number of ingredients in the gel or slicks categories tend to increase toxicity, while increases in the number of ingredients in the unspecified category tend to decrease toxicity. While these effects are statistically important, they are not large in magnitude, calling their economic significance into question.

The fourth set of results we obtain are related to the firms involved in the wells. We observe that operator I is associated with significantly greater use of toxics, while operator II is not; in both cases, the comparison is against III and IV. The conclusion we draw is that operator I exhibits a greater tendency to use toxic ingredients than do its rivals. Among the service companies, we see that service company B is associated with more toxic ingredients than either service company A or C. These differences are both statistically and economically important – so we infer that firms differ in important ways.

Fifth, the number of toxics associated with wells completed in 2012 and 2013 is significantly larger than wells completed in 2010; there is no difference between wells completed in 2011 and 2010. This observation casts doubt on the view that the regulation enacted in 2010 induced firms to become more cautious about using toxic ingredients in their wells.Footnote ¹⁸