A. Historical setting

In 1986, following the three-tier system of alcohol distribution that became prominent after the repeal of the Eighteenth Amendment, California passed the first “Reciprocity” legislation, which would come to define the progression of wine distribution for the next 20 years. Counterintuitive on its face, this legislation leveraged the size of California by preventing California consumers from purchasing wine directly from a winery located in any other state that restricted the direct purchase of California wine. The ultimate aim of this move was to break down the restrictions across the country that prevented DTC (Riekhof and Sykuta, Reference Riekhof and Sykuta2005). For instance, if a state like Ohio did not allow its residents to purchase wine directly from California wineries, then this “Reciprocity” legislation would prevent Californians from directly purchasing Ohio wine. In 1986, the population of California was nearly 3x larger than the population of Ohio, and so legislation was aimed to disproportionately affect non-California wineries. The hope was that other states would change their DTC policies, allowing California wineries to directly ship to their consumers.

Between 1986 and 2001, 12 other states followed California's lead by passing legislation to become Reciprocal States.Footnote ^4, Footnote ⁵ Within reciprocal agreements, the additional effort to distribute directly to consumers was minimal: wineries were not required to obtain a license or pay additional taxes to ship to consumers within the group of 13 Reciprocal States. However, wineries located in states outside of these 13 were not permitted to ship directly to consumers in these 13 states.

Between 1996 and 2003, nine other states adopted a form of DTC shipping laws whereby wineries from all states could directly ship to consumers in their state, given a (typically) small permit fee and taxes.Footnote ⁶ The key differences between the Permit State DTC legislation and the Reciprocal State legislation were: First and primarily, the DTC restrictions removed in Permit States were extended to wineries in every state. The same was not true for Reciprocal States, where only wineries within another Reciprocal State had the privilege of direct shipping to their consumers. Second, Permit States required that wineries obtain a permit in order to ship to consumers in their state. Third, while wineries and consumers located in Reciprocal States were not required to pay sales or excise taxes from wine purchased within the 13-state agreement, generally, Permit States required taxes to be paid by the winery and consumer (both in-state and out-of-state).

A third kind of DTC that existed prior to the Granholm decision was a group of states that disallowed out-of-state wineries from shipping directly to their consumers but did allow in-state wineries to do so.Footnote ⁷ Wineries from these states were thus able to sell to consumers in Permit States and to consumers in their own state, but could not sell to consumers in Reciprocal States or in states that outlawed DTC all together. This genre of policy ultimately formed the most visible basis of the case argued at the Supreme Court.

In May 2005, the Supreme Court of the United States ruled that reciprocal agreements and in-state protection were in violation of the Commerce Clause of the Constitution. States were permitted to regulate the sale of alcohol, but they had to apply the regulations equally. In the majority opinion of Granholm v. Heald, Justice Anthony Kennedy opined, “…the object and effect of the laws are the same: to allow in-state wineries to sell wine directly to consumers in that State but to prohibit out- of-state wineries from doing so, or, at the least, to make direct sales impractical from an economic standpoint.”

Some Reciprocal States transitioned quickly to become a Permit State,Footnote ⁸ while other formerly Reciprocal States transitioned much more slowly.Footnote ⁹ Table 1 shows the number of states that allowed DTC from at least one other state over time. Column (1) of Table 1 shows the progression over time of Reciprocal States transitioning to become Permit States, from 13 in 2002 to 0 in 2011. The impetus for this change was the Granholm decision in 2005. Column (2) shows the number of Permit States, which grows both from transitioned Reciprocal States and from an increase in states that passed DTC legislation.Footnote ¹⁰

Table 1. Number of states that allowed DTC from wineries in at least one other state

Notes: Reciprocal States refer to the 13 states that allowed DTC among themselves. Non-Reciprocal States refer to states that allowed some form of limited DTC (e.g., a license or permit was required). Post Granholm decision, all Reciprocal States eventually transitioned to becoming permit states.

On one hand, there is evidence that the sale of alcohol over the internet may increase access to minors (Williams and Ribisl, Reference Williams and Ribisl2012), and so states may be operating to reduce public health risk;Footnote ¹¹ on the other hand, the regulation that wineries must ship through a distributor may act as a means of market power for large wineries, at least in theory. Indeed, Mullins (Reference Mullins2009) found that only a minority of small wineries sell wine through a distributor, while the majority of large wineries have distributor representation.

B. The prevalence of DTC

While DTC is not a new phenomenon, its practice is increasingly valuable for wineries, and its popularity continues to grow among consumers as travel and shipping become easier and less expensive.Footnote ¹² The Federal Trade Commission (FTC) released a report in 2003 used in the Supreme Court case that stated that DTC transactions made up 3% of total wine sales between 1994–1999 (FTC, 2003). More recent estimations put the percent of the total $30 B of off-premises wine sold through DTC channels at 10% (Savos ShipCompliant, Reference ShipCompliant2020). This demonstrates that while DTC is a relatively minor part of total wine sales, it is still a substantial and growing portion of the wine market.

A. Policy endogeneity

Threats to causality in this model rely on the possibility that $cov( {TotalMarket, \;\epsilon } ) \ne 0$. This might arise if there exists some unobserved, time-varying characteristic that influences the Total Market available to a winery in state s. As noted earlier, a commonly discussed scenario of policy endogeneity is related to the phenomenon called the “Ashenfelter Dip” (Ashenfelter, Reference Ashenfelter1978). If a state's policy were to change simultaneously with, or in response to, changes in the wine market, then the “Ashenfelter Dip” would be an important critique and consideration.

Because the total potential market is comprised of the sum of the in-state and out-of-state adult populations, an expansion of the market occurs both through decreased DTC restrictions by state s itself—that is, increasing the In-State portion of Equation (1)—or through decreased regulations by other states—that is, increasing the Out-of-State portion of Equation (1). A congruent effect of the “Ashenfelter Dip” would be that changes in state DTC policy would result from changes in the industry. I argue that changes in the in-state potential market may be related to an unobserved characteristic of the state, but that changes in the out-of-state potential market are implausibly related to unobserved characteristics of the state's wine industry. To render this latter statement plausible would suggest, for example, that the removal of DTC restrictions in California are related to an idiosyncrasy of Ohio's wine industry. As a result, for robustness, I also tried a similar model where TotalMarket_s,t is replaced only with the Out-of-State portion of the Total Market equation. This specification removes all possibility of policy endogeneity and yields the unbiased, causal effect of DTC restrictions on the wine labor market. Moreover, because the results from restricting the full potential market to the out-of-state potential market are unperturbed,Footnote ¹⁶ there is strong suggestive evidence that the full model gives a causal estimate of the effect of market expansion due to decreased DTC restrictions on winery establishment count and employment.

B. Winery size distribution

Next, I delve into a particular hypothesis of this article, namely, that decreased DTC restrictions disproportionately benefit small- and medium-sized wineries, relative to large-sized wineries. This hypothesis draws on the observations that small wineries are less likely to have distributor representation (Mullins, Reference Mullins2009), in part due to regulation (Santiago and Sykuta, Reference Santiago and Sykuta2016), and that DTC comprises a large portion of a small winery's revenue (Peden, White, and McMillan, Reference Peden, White and McMillan2018). This hypothesis is in line with theoretical expectations outlined later.

Because the effect of market expansion via the removal of DTC restrictions may be different for small-, medium-, and large-sized wineries, I model the expansion of markets on the number of wineries of size j in a given state s at time t. Using a difference-in-difference-style framework, which supposes that the trends in the log-number of firms for each group are similar and would have remained similar without any change in market size, I operationalize the model as follows:

(3)$$\eqalign{& ln( y ) _{\,jst} = {\rm \alpha } + {\rm \beta }_1M_{s, t} + {\rm \beta }_2( {small\;\ast\; M} ) _{\,j, s, t} + \beta _3( {med\;\ast\; M} ) _{\,j, s, t} + {\boldsymbol X}\beta + \rho _{\,j, s} + \tau _t + \epsilon _{\,j, s, t},}$$

where M is the size of the potential market for a winery in state s at time t, as calculated by Equation (1); and y is one of the outcome variables described earlier. The interaction terms can be interpreted as the “premium” that each size category derives from an increase in the size of the market available to a winery in that state relative to large wineries. Tangibly, the coefficients of interest in this equation are β ₂, and β ₃, which will reveal any potential differences in how the number of wineries in state s for small and medium wineries relative to large wineries differs as the market expands as a result of the removal of restrictions preventing DTC. As in the previous set of models, the time fixed effects remove the effect of a secular, national trend in the number of wineries; the state-by-size fixed effects capture any time-invariant idiosyncratic differences of a particular size category in a given state.

A. Size distribution of wineries

Firms chase economic profits. Conventional economic theory says that where economic profit exists, firms will enter in an effort to extract some of the available profit. In a perfectly competitive market, firms enter until economic profit is driven down to zero, and the firm can cover the cost of its factors of production but is unable to extract additional profit. However, when market barriers exist, the flow into markets is frustrated, and the market is characterized by monopolistic power, that is, the long-run ability to extract positive economic profit. In the case of DTC regulations, which require that wineries sell wine through a distribution channel, this is a clear barrier to entry.

Because large wineries are much more likely to be represented by a distributor than are small wineries (Mullins, Reference Mullins2009), and because DTC sales comprise a larger portion of a small winery's sales than those of a large winery (Peden, White, and McMillan, Reference Peden, White and McMillan2018), we expect that regulations that require wine to be purchased through the three-tier system will disproportionately harm small wineries. Hence, the removal of DTC regulations will disproportionately benefit small wineries. Moreover, if we assume that the probability of distributor representation is positively associated with winery size, then we would expect that the benefits from reduced DTC restrictions are inversely related to winery size.

This expectation should manifest itself as an increase in the number of wineries. As the number of potential customers expands for a winery in a given state, the opportunity for profit will drive the number of wineries up, and in particular in the strongest way for small wineries that lack distributor representation. These observations form the basis for my empirical model.

B. Wineries from established states

Another potential effect is that wineries from established states may prosper disproportionately as a result of the removal of DTC restrictions. A state that has very little production initially may be unlikely to see much immediate benefit from DTC since the reputation of the wineries has not progressed to a large consumer base. Past research has documented a perceived stigma against wines from unestablished states (Costanigro, McCluskey, and Goemans, Reference Costanigro, McCluskey, Goemans, Ashenfelter, Gergaud, Storchmann and Ziemba2010). Instead, when a state with many established wineries, like California, gains access to a large [potential] market, one might expect its wineries to flourish. Indeed, there is some evidence that a more favorable market may not benefit those with deficient reputations in the same way that those with already-positive reputations can withstand detrimental changes in the environment (Levin, Reference Levin2009).

To illustrate this point: Several states maintain good, detailed records of DTC licenses issued to wineries. For the State of Ohio, since the inception of the “S” permitFootnote ¹⁷ in 2007 until summer 2022, there have been a total of 1,494 licenses issued. Of these 1,494 licenses, 887 (i.e., 60%) were issued to California wineries to ship directly to Ohio (State of Ohio Department of Commerce, 2020). In 2007 alone, when the license first began, 133 “S” permits were issued to California wineries (67%), 19 Ohio wineries (10%), and 48 wineries from other states. As of 2020, in Wisconsin, California wineries occupy nearly 53% of all permits issued. While California is by far the largest producer of wine, its wineries also appear to be taking advantage of expanded wine markets, whereas other states may not be, at least initially.

On the other hand, as states allow for increased direct commerce, wineries from states that are not traditional producers may catch up to wineries from established wine states.Footnote ¹⁸ This traction might be small in absolute terms, but the rate of change for a small-producing state could still be high. Peden, White, and McMillan (Reference Peden, White and McMillan2018) show in a Silicon Valley Bank report that DTC in 2018 was even more important for non-traditional wine states, as it constituted a larger part of the winery's revenue. This fact provides some evidence that wineries in non-traditional states are taking advantage of reduced DTC restrictions.

A. Production data

Production data is retrieved from Wine Statistical Releases, produced monthly by the Alcohol and Tobacco Tax and Trade Bureau (TTB). Table 2 shows that overall wine production increased by 37% between 2002–2005 and 2014–2017, but that the share of production among the top 10 wine-producing states, top 3 wine-producing states, and California declined by 5–8 percentage points over the same years.Footnote ¹⁹

B. DTC market

The market size variable is generated as discussed in the Empirical Model section to reflect the potential consumer base that is afforded to each winery when DTC restrictions are removed.Footnote ²⁰ This, then, allows for differentiation between the adoption of DTC by a state like South Dakota and a state like Pennsylvania.Footnote ²¹ Furthermore, it allows for an intuitive interpretation of the differences between Reciprocal States (whose distribution populations were comprised of the populations of the 13 states plus the populations of the Permit DTC States); a state that could distribute only to Permit States; and a state that only allowed in-state wineries to distribute to consumers (prior to Granholm). Ultimately, the DTC market variable gives a continuous representation of the total market, internal and external, to which a winery from a given state could distribute, which, when interacted with, allows me to distinguish between California gaining access to South Dakota and South Dakota gaining access to California. Table 2 shows that the DTC market expanded most rapidly for non-formerly Reciprocal States. While formerly Reciprocal States saw an increase in the DTC market of over 100%, non-formerly Reciprocal States saw a much larger increase of over 600%.

Figure 1 expands on this observation in Table 2, by showing the comparison of the mean DTC market between formerly Reciprocal and non-Reciprocal States, defined in 2004, as a time series. As expected, non-Reciprocal States see the bulk of the growth in the DTC market, as they were eventually granted access to California and 12 other states that were not available to them prior to Granholm v. Heald.

Figure 1. Average potential market (in millions) for formerly-reciprocal and Non-Reciprocal States.

C. Employment and establishment count

The raw data for the employment and establishment count variables come from the CBP, which is produced annually by the U.S. Census Bureau (2002–2017b). The dataset provides establishment count and employment data by NAICS and by winery size. While the raw data provide an accurate establishment count by winery size regardless, CBP redacts information about employment in instances where there are very few wineries of a certain size category in a state. For a smaller NAICS industry, like wineries, this occurs more frequently than in larger industries. In situations where this occurred, employee count was imputed by taking the midpoint of the range of employees under which the winery was classified, and multiplying it by the number of wineries in that size class.Footnote ²² Both Establishment Count and Employee Count are used as dependent variables to examine the effect of increased DTC market on their rates of change. Because CBP reports nine size categories, the nine categories are collapsed into three groups: small wineries, medium wineries, and large wineries.

Table 2 also shows that the growth in the number of small wineries (1–9 employees) and medium wineries (10–99 employees) substantially outpaced the growth in large wineries (100+ employees). Small and medium wineries saw faster rates of change in employment, too, as compared to large wineries. Between 2002–2005 and 2014–2017, small and medium wineries saw growth rates of 159 and 117%, respectively, for establishment count, versus 1% growth in large wineries. Small wineries also saw 152% growth in total employment, while medium wineries saw a 105% change in employee count. Large wineries saw only a 37% increase in employment. This is the total measure of how many employees worked in small, medium, and large wineries, which can be an indicator of either (or both) an increase in the number of establishments or more employees hired. However, both the share of employment and the share of establishments in the top 10 wine-producing states in 2002, the top 3 wine-producing states in 2002, and California saw declines of approximately 7–14 percentage points between 2002–2005 and 2014–2017.

D. Controls

As I argued earlier, provided that the Conditional Independence Assumption is met, the total potential market variable should be identified without controls. However, I also control for a number of in-state time-variant factors. The results vary little. In particular, I add controls for numerous types of tax revenue collected by the state in a given year, as reported by the U.S. Census Bureau (2002–2017a) dataset;Footnote ²³ state unemployment through the Federal Reserve of Philadelphia (n.d.);Footnote ²⁴ the political party of the state governor; and DTC permit fees.

A. Total wineries and total employment

Table 3 demonstrates the results of the impact of decreased regulations on the number of wineries, and Table 4 shows the results for employment.

Table 3. Aggregate effect of DTC market expansion on winery establishment count

Notes: Clustered standard errors: *p < 0.05, **p < 0.01, and ***p < 0.001.

The dependent variable for all models is log(establishment count + 1). Total Market (in millions) refers to the sum of the adult population in states where a winery could potentially ship directly. For robustness, External Market in model (6) excludes the population from the state. The dependent variable for (3) is log(establishment count + 1) for NAICS 10 (all industries) less wineries. Permit fees, coincident indexes, state governors’ political parties, year and state fixed effects, and various total state tax collections are included in models (2)–(7) as controls. For robustness, models (4) and (5) verify the parallel trend assumption. Column (7) includes the count of out-of-state wineries. Standard errors are clustered at the state level.

Table 4. Aggregate effect of DTC market expansion on winery employment

Notes: Clustered standard errors: *p < 0.05, **p < 0.01, and ***p < 0.001.

The dependent variable for all models is log(employment + 1). Total Market (in millions) refers to the sum of the adult population from states where a winery could potentially ship directly. For robustness, External Market in model (6) excludes the population from the state. The dependent variable for (3) is log(employment + 1) for NAICS 10 (all industries) less wineries. Permit fees, coincident indexes, state governors’ political parties, year and state fixed effects, and various total state tax collections are included in models (2)–(6) as controls. Models (4) and (5) consider lags in market expansion. Standard errors are clustered at the state level.

Analyzing the results from Table 3, we see that an expansion of the potential market by 10 million people is associated with approximately a 3.5% increase in the number of wineries.Footnote ²⁵ Column (1) shows this relationship without any controls. This coefficient estimate is remarkably consistent with the estimate from our preferred specification in Column (2), which adds controls. The estimate is also similar to that in Column (6), which includes only the out-of-state portion of the potential market as described in Equation (1). Shown in Table 2, the average state Adult Population between 2014–2017 was about 5 million people. Thus, using my estimate of a 3.5% increase in the number of wineries per 10 million people, this estimate yields upwards of a 2% increase in wineries for each additional state that removes DTC restrictions.

Because this design is a quasi-difference in difference, I include several robustness checks in Columns (3)–(7). Column (3) examines the effect of expanded markets on the total number of non-winery establishments and reveals a null effect, as expected. This implies that the expanded markets primarily affected the wine industry, with a null effect on the number of establishments overall. Columns (3) and (5) follow the suggestions made by Angrist and Pischke (Reference Angrist and Pischke2008, pp. 177–178) to analyze any potential pre-treatment trends through the notion of Granger Causality (Granger, Reference Granger1969), in order to satisfy the parallel trends assumption of the model. Given the atypical formulation of this quasi-difference in difference design, a typical event study analysis is unclear conceptually, and so I use parametric techniques to validate the parallel trends assumption. The intuition in this set of robustness checks is to assess the extent to which a marginal increase in a future potential market predicts establishment count in time t. By including lead terms in the total market, we should expect a near-zero and statistically insignificant estimate of the lead terms. Indeed, both Columns (4) and (5) show this, suggesting that there are parallel trends prior to treatment. Column (7) addresses the potential concern that the number of competitors from other states may matter. I include the number of out-of-state wineries as a control variable, and the coefficient estimate remains stable. Because specification (7) is similar to the preferred specification, this provides evidence that competition does not moderate the effect of expanded markets from decreased DTC restrictions. Further, it provides further evidence of the exogeneity of the potential market variable, as discussed in Section III.A.

Moving to Table 4, we see a very different story for employment. Despite a statistically significant 3.7% increase in the number of firms per 10 million new potential customers, the results from the employment models reveal, at best, delayed effects. I find no statistical effect of market expansion via decreased DTC regulations on employment for wineries, as shown in Columns (1) and (2). While the point estimate of Column (6) aligns with the point estimate of the effect in Column (2), again suggesting that unobserved state characteristics are not associated with the market size variable, the effect for both models is null. This suggests that an expansion of the potential consumer base does not increase employment at wineries overall. Columns (4)–(5) follow Angrist and Pischke (Reference Angrist and Pischke2008) by including lagged terms. The results from these models suggest that employment increases may be delayed. In particular, after 1 year, the impact of decreased restrictions on employment is approximately 11% for an increase of 10 million new potential customers; after 2 years, decreased regulation is associated with approximately an 8% increase in employment. These results suggest that labor markets are slower to react than firm entry.

B. The size distribution of wineries

Now, I turn to my second set of models, by examining the effect of expanded potential wine markets through decreased DTC regulation on the size distribution of wineries. Table 5 examines the effect of decreased regulation on the size distribution of the count of wineries. Models (1) and (5) provide the most direct answer to the question, using log and inverse hyperbolic sine of establishment count as the dependent variables, respectively.Footnote ²⁶ Column (1) seems to substantiate part of the Theoretical Expectations that I outlined earlier, namely, that smaller wineries will benefit to a greater degree than large wineries. I find that the effect of expanded potential DTCs has a negative effect on large wineries, while small and medium wineries appear to benefit. For an increase of 10 million potential customers, small wineries grow by nearly 6%-points more than large wineries, or by over 1%-point more than medium wineries. For an increase of 10 million potential consumers, large wineries, on the other hand, decline by nearly 4%.

Table 5. Effect of decreased DTC regulations on winery size count distribution

Notes: Clustered standard errors: *p < 0.05, **p < 0.01, and ***p < 0.001.

The dependent variable for models (1)–(4) is log(establishment count + 1). Total Market (in millions) refers to the sum of the adult population from states where a winery could potentially ship directly, and is abbreviated as TM for space. Reciprocal State is an indicator for whether the state was a Reciprocal State prior to Granholm v. Heald. Model (5) uses an inverse hyperbolic sine in place of log, for robustness. Permit fees, coincident indexes, state governors’ political parties, year and size-by-state fixed effects, and various total state tax collections are included in each model. Standard errors are clustered at the state, size category level.

Models (2)–(4) examine the second set of theoretical expectations, namely the impact of expanded wine markets, via decreased DTC regulation, on established wine states. In Columns (2)–(4), I interact the potential market with winery size category and an indicator of whether the state was one of the top 10 wine-producing states in 2002, a formerly-Reciprocal State, California, respectively. The results from Column (4) show that, while small and medium wineries are still better off due to the decrease in DTC regulations, the positive effect appears to be largely experienced by wineries located outside of California. Moreover, I find no evidence that the number of wineries grew faster in formerly-Reciprocal States than if located in other states, which offers some contrast to past research that found a perceived stigma against non-traditional wine states (Costanigro, McCluskey, and Goemans, Reference Costanigro, McCluskey, Goemans, Ashenfelter, Gergaud, Storchmann and Ziemba2010). My results also show that the wine market in non-traditional wine states tends to grow no faster than the market in established wine states.

Turning to Table 6, I again find largely null effects of decreased DTC regulations on employment. I do find that for an increase in the total potential market of 10 million people, small wineries in California have a slightly positive premium compared to large wineries in California. Conversely, I find slightly negative growth in employment at small, California wineries relative to their non-California counterparts.

Table 6. Effect of decreased DTC regulations on winery employment

Notes: Clustered standard errors: *p < 0.05, **p < 0.01, and ***p < 0.001.

The dependent variable for models (1)–(4) is log(employment + 1). Total Market (in millions) refers to the sum of the adult population from states where a winery could potentially ship directly, and is abbreviated as TM for space. Reciprocal State is an indicator for whether the state was a Reciprocal State prior to Granholm v. Heald. Model (5) uses an inverse hyperbolic sine in place of log, for robustness. Permit fees, coincident indexes, state governors’ political parties, year and size-by-state fixed effects, and various total state tax collections are included in each model. Standard errors are clustered at the state, size category level.