Professional Documents
Culture Documents
In July 2016, Vermont became the first U.S. state to require mandatory labeling of foods containing
genetically engineered (GE) ingredients. The introduction of the Vermont law serves as a quasi-
natural experiment on the economic effects of mandatory GE labeling. We investigate the market re-
sponse in the U.S. sugar market. Almost all beet sugar is GE, while cane sugar is GE-free. Prior to
2016, cane and beet sugar were regarded as homogenous. However, in mid 2016, refined cane sugar
began selling at a premium over refined beet sugar. We find the mandatory labeling initiative gener-
ated about a 13% price discount for beet sugar and a premium of about 1% for cane. Food manufac-
turers’ concerns over mandatory labeling caused them to switch inputs. This resulted in a
redistribution of welfare in the U.S. sugar industry.
Key words: GMO labeling, Vermont Act 120, National Bioengineered Food Disclosure Standard,
sugar market.
JEL codes: Q13, Q18.
“Many food companies have decided engineered (GE) ingredients is one such exam-
to label their products as non-GMO. ple of this phenomenon.1
And because practically all sugar In May 2014, Vermont became the first
beets in the U.S. are genetically mod- (and only) U.S. state to pass mandatory label-
ified, those food products are now ing rules for GE foods sold for consumption
using sugar derived from sugar at home.2 The law (Act 120) took effect on
cane. . .” July 1, 2016, but was nullified by federal legis-
Dan Charles, National Public lation later that month. Penalties and require-
Radio (May 12, 2016) ments under the Vermont law were harsh.
Consumers could sue for violations of the la-
Legislation is not made in a vacuum. Markets beling requirements “without needing to
evolve even as policy debate transpires, and demonstrate any specific damage occurred as
markets anticipate policy changes. Once a pol-
a result of the alleged violation,” (Bovay
icy outcome becomes sufficiently likely, pro-
and Alston 2018). Manufacturers of improp-
ducers adjust to minimize costs and maximize
benefits. But these adjustments can be costly to erly labeled products sold in Vermont would
reverse even if the policy outcome never comes be fined $1,000 per day, per product.
to pass or is later quashed. Segmentation of the About two weeks after the Vermont law
U.S. sugar market in response to mandatory la- took effect, U.S. Congress responded with
beling legislation for food containing genetically the National Bioengineered Food Disclosure
Standard (NBFDS), which pre-empts state-
1
GE technology is defined here as in Van Eenennaam
Colin A. Carter is a distinguished professor of the Department of et al. (2014): “Genetic engineering (GE) can be defined as the
Agricultural and Resource Economics, University of California, manipulation of an organism’s genes by introducing, eliminating,
Davis and the Giannini Foundation of Agricultural Economics, or rearranging specific genes using the methods of modern mo-
Davis. K. Aleks Schaefer is a lecturer of the Royal Veterinary lecular biology, particularly those techniques referred to as re-
College, University of London, UK. The authors would like to combinant deoxyribonucleic acid (rDNA) techniques.”
2
thank editor Timothy Richards and three anonymous referees Exempted from labeling requirements were alcohol, food
for comments that greatly improved the article, and Michael J. served in restaurants, ready-to-eat foods, foods derived from ani-
McConnell for his help in obtaining the U.S. sugar price data. mals raised with GE feed, and foods manufactured with GE
Correspondence may be sent to: aschaefer@rvc.ac.uk. processing aids, such as enzymes (Bovay and Alston 2018).
level labeling initiatives and imposes federal products; thus, understanding the effects of
disclosure requirements for some GE- labeling laws on the sugar market has broad
containing foods but allows that the disclo- relevance across a wide range of consumer
The second question is one of impact—if sorting mechanism for consumers (Crespi
labeling requirements caused the price diver- and Marette 2003; Fulton and
gence, what would refined sugar prices have Giannakas 2004) and as a signal about the
been in the absence of such legislation? relative safety of GE foods (Artuso
Using a standard regime-switching regression 2003; Lusk and Rozan 2008). The consumer
model, we find that the Vermont law induced impacts and potential price differences be-
a break in the price relationship between tween GE and non-GE ingredients also affect
U.S. and world sugar prices. This price pre- manufacturers’ input and supply decisions
mium has persisted under the NBFDS. (Carter and Gruère 2003b).
Relative to what prices would have been in
the absence of legislation—GE labeling Firm-Level Decision
requirements generated an average premium
for cane sugar of approximately 1% and a The introduction of the Vermont law forced
discount for beet sugar of around 13%. In the manufacturers to make a binary choice: (a)
12-month period immediately following the maintain current product formulation and la-
July 2016 legislation, mandatory GE labeling bel or (b) switch to non-GE inputs to avoid
requirements generated a $40 million wind- labeling. The profit-maximizing strategy is a
fall for U.S. cane refiners and cost U.S. beet function of the share of consumers willing to
processors approximately $400 million. buy GE-labeled foods, the cost of switching
from GE- to non-GE inputs, and the cost of
compliance with the labeling policy (Carter
Economics of GE Food Labeling Laws & and Gruère 2003b). Consumers may interpret
Input Mix a mandatory label as a warning that GE foods
pose a food safety threat or are of poor qual-
Economists have studied the drivers and in- ity. If so, some consumers previously unwill-
ternational differences in GE labeling ing to pay a premium for non-GE may pay to
requirements (Carter and Gruère 2003a; avoid GE as a result of the policy. Moreover,
Gruère, Carter, and Farzin 2009; Vigani for most food products reformulation to-
and Olper 2013; McCluskey, Wesseler, wards non-GE inputs is unlikely to have a
and Winfree 2018) and the impact of manda- substantial effect on the final food price. GE
tory labels on consumer and producer behav- ingredients are typically used in highly proc-
ior. From the consumer perspective, essed products and represent a small share of
mandatory GE labels function both as a the food dollar (Cowan 2010). These factors
4 August 2018 Amer. J. Agr. Econ.
push the food manufacturer to convert to the decision to reformulate away from GE
non-GE inputs to avoid labeling. ingredients and towards non-GE ingredients
A similar choice exists for retailers sends a message of quality upgrading to
depicts the residual demand for refined sugar processors is ðPf PBeet ÞQBeet . The introduc-
(net of TRQ and Mexican imports) by U.S. tion of GE labeling requirements also results
food manufacturers and other sugar users. in a deadweight loss equal to the sum of tri-
Equilibrium occurs where market demand angle k in panel (a) and triangle j in panel
for sugar equals market supply, shown by the (b).
intersection of schedules SSugar and DSugar in
panel (c). In this equilibrium, QSugar units of
sugar were sold, QBeet of which were derived Methodology & Preliminary Analysis
from beets and QCane of which were derived
from cane (where QBeet þ QCane ¼ QSugar ). In this section, we examine whether the theo-
All sugar was sold at price Pf. retical and anecdotal evidence regarding the
Enactment of a mandatory labeling law seg- relationship between GE labeling laws and
mented the demand schedule for refined sugar. U.S. sugar prices is supported by empirics.
A portion of the market, including food manu- We first use a structural break test to deter-
facturers who reformulate towards non-GE, mine whether the timing of the break in cane
was now willing to pay a premium to purchase relative to beet prices is consistent with the
sugar derived from cane. Demand for GE-free legislative and policy timeline. We then use a
cane sugar is depicted as schedule DGEfree in regime-switching regression model to mea-
panel (b). Those who do not reformulate or are sure the historical relationship between U.S.
unwilling to pay a premium for cane will con- and world refined sugar prices prior to the
tinue to source from beet (depicted as schedule observed break. Finally, we compare pre-
DBeet in panel (a)). and post-structural-break prices for refined
Under a mandatory labeling policy, equi- sugar to measure the effects of the labeling
librium occurs when the demand for GE-free legislation. Model robustness and sensitivity
sugar is equal to the supply of cane sugar and are also considered.
the demand for sugar users unwilling to pay a Our data includes monthly observations of
premium aligns with the supply of beet sugar. U.S. raw and refined cane sugar, refined beet
In figure 2, this equilibrium is represented by sugar, and world raw and refined sugar prices
the intersection of schedules DGEfree and from January 2011 to December 2017. The
SCane in panel (b) and the intersection of U.S. raw cane price is the Intercontinental
schedules DBeet and SBeet in panel (a). Exchange (ICE) Sugar No. 16 nearby futures
The law drives a wedge between the price contract.5 The world raw price is the nearby
of non-GE and GE inputs. Processors of non- price for the ICE No. 11 contract.6 There is
GE inputs receive a premium for their prod-
ucts, while processors of GE inputs receive 5
Monthly prices were obtained from table 4 of the ERS Sugar
lower prices. The premium to cane refiners is and Sweeteners Yearbook. “Nearby” refers to the contract with
PCane Pf and the discount to beet process- the closest settlement date. The ICE No. 16 contract specifies
ors is Pf PBeet . The policy results in trans- that 112,000 pounds of raw cane sugar be physically delivered to
one of five U.S. refinery ports: New York, Baltimore, Galveston,
fers of economic welfare away from beet New Orleans, or Savannah. Delivery months are January, March,
processors and towards cane refiners. The to- May, July, September, and November.
6
Monthly prices obtained from table 3b of the ERS Sugar and
tal windfall to cane refiners is Sweeteners Yearbook. The No. 11 contract specifies delivery of
ðPCane Pf ÞQCane . The total loss to beet 112,000 pounds of raw cane sugar in delivery months March,
6 August 2018 Amer. J. Agr. Econ.
no futures market for refined sugar in the Table 2. VECM Cointegration Results (Jan.
United States. We use the average monthly 2011–Dec. 2015)
spot price for refined cane sugar and the aver-
age monthly spot price for refined beet sugar Series Obs. v2 P-Value Conclusion
published by the Milling & Baking Cane; Beet 58 2688.647 0.00 Cointegrated
Magazine.7 The world refined price is the Cane; World 58 6.872 0.01 Cointegrated
nearby price for the ICE No. 5 London Daily Beet; World 58 6.769 0.01 Cointegrated
futures contract for white (i.e., refined) sugar
Note: All prices are monthly averages and are specified in natural logarith-
free-on-board in Europe.8 Deliveries under mic form.
the London No. 5 contract can be either re-
fined beet or cane sugar.
The start date is purposefully chosen as the world market would be fully passed
January 2011 to limit the analysis to the rele- through to the U.S. market (and vice versa).
vant market environment. The volume of A preliminary question is whether U.S. and
Mexican sugar exports to the U.S. increased
world sugar prices move together in this pol-
substantially in fiscal year (FY) 2011 and has
icy environment. To formally test whether
remained high since then (United States
U.S. prices and world prices are cointegrated,
International Trade Commission
we construct a vector-error correction model
[USITC] 2015). This change in export vol-
(Engle and Granger 1987). To verify that re-
umes likely affected the relationship between
fined cane and beet sugar prices were cointe-
U.S. and world prices, on which our regres-
sion analysis hinges. grated prior to the introduction of the
Table 1 presents Augmented Dicky-Fuller mandatory labeling law, we measure the coin-
(ADF) test statistics for monthly U.S. and tegrating relationship between Jan. 2011–
world sugar prices from Jan. 2011 to Dec. Dec. 2015. We remove later periods because
2017 (Dickey and Fuller 1979). The third col- they are potentially affected by the manda-
umn reports the corresponding MacKinnon tory labeling requirements. The analysis indi-
approximate p-value. As shown in table 1, we cates a statistically significant long-run
fail to reject the null hypothesis of non- cointegration relationship between all three
stationarity for all price series. In other price series (results are reported in table 2).
words, U.S. and world sugar prices follow a Prior to the imposition of GE labeling laws,
random walk. U.S. cane and beet refined sugar prices were
Under U.S. sugar policy, all sugar imports cointegrated with each other and each series
are subject to volumetric restrictions and all was cointegrated with world refined prices at
U.S. production is subject to marketing quo- the 99% level of confidence.
tas. In this setting, it is unlikely that shocks to
ð1Þ Pct ¼ a þ bPbt þ et : in logs. The two tests are generally consistent.
July 2016 is the most likely date for the break
Variable Pct represents the U.S. price for re- in the Wald specification and the second-
fined cane sugar at time t, and Pbt is the corre- most-likely date in the L.R. specification.
sponding U.S. price for refined beet sugar. Similarly, May 2016 is the most likely date for
We consider Supremum Wald and the break in the L.R. specification and the
Likelihood Ratio (L.R.) tests for a structural second-most-likely in the Wald specification.
break at an unknown break date using a sym- Interestingly, the figure shows a substantial in-
metric, 15% sample trim (Quandt 1960; Kim crease in the both the Wald and L.R. statistic
and Siegmund 1989; Andrews 1993). for May 2014—the month the Vermont
We conduct the analysis with variables speci- Labeling Law was passed.
fied in both levels and natural logarithmic These findings strongly support the conclu-
form. sion that the break in sugar prices was the re-
Table 3 reports the results of the structural sult of the GE labeling initiatives. Food
break tests. All tests reject the hypothesis of manufacturers faced with the pending
no structural break with 99.99% confidence. Vermont legislation (and other state laws on
The Wald tests identify July 2016—the month the horizon) initially waited to convert to
the Vermont GE Labeling Law took effect GE-free ingredients in hopes that federal leg-
and the NBFDS was enacted—as the most islation would invalidate mandatory labeling
likely date when the structural break oc- rules. When the Vermont law became immi-
curred. The Likelihood Ratio tests identify nent, food intermediaries reformulated away
the break date two months earlier, in May from GE ingredients (beet sugar) towards
2016. non-GE ingredients (cane sugar), resulting in
Figure 3 plots Wald and L.R. statistics for a price premium for cane and a discount for
each candidate break date for the specification beet.
8 August 2018 Amer. J. Agr. Econ.
There is still the possibility that the timing post-July 2016 correlations for refined-to-raw
of the GE labeling legislation and the price sugar have important implications for the dis-
divergence could have been coincidental. For tributional impacts of the policy. As noted in
example, the breakdown in the cane-to-beet the introduction, sugarcane is first processed
cointegrating relationship could have been into raw sugar and then sent to refiners for
caused by non-U.S.-centric factors or issues further processing into refined sugar. In con-
unique to the United States but occurring trast, sugar produced from sugarbeets does
elsewhere along the sugar supply chain. To not have a “raw” stage; processing is a single,
conclusively attribute the cane-to-beet price continuous process from beet to refined
break to mandatory labeling requirements, sugar. The fact that the price impact does not
one must consider these alternative explana- appear to have been transmitted to the raw
tions. For clarity and brevity, we focus on the stage of cane production suggests that—over
July 2016 candidate break in the analysis that the period of analysis—any benefits of the
follows. We have conducted a corresponding policy were most captured by cane refiners.
analysis for the May 2016 candidate break. The world market did not experience a diver-
Because the difference involves only two gence in the relationship between raw (cane)
data points from a much larger sample, point sugar prices and the price for refined sugar
estimates between the two breakpoints are (which may be filled with either cane or beet
virtually identical for all analyses that we sugar) in July 2016. As shown in the final two
have conducted. All findings and discussion columns of table 4, the correlation between
that follow are robust to the use of the May world raw and refined sugar prices was 0.99
2016 break date. before July 2016 and 0.98 afterward. The lo-
Table 4 reports the correlation between cus of the shock is the U.S. refined sugar
U.S. and world sugar prices (again in natural market.
logarithmic form) prior to and after July
2016. As shown in column 1, the U.S. refined Price Impact
cane price and the U.S. refined beet price had We now turn to the question of impact—
a correlation of 1.00 prior to July 2016. In what would U.S. refined sugar prices have
other words, a shock to the cane price was been in the absence of GE labeling legisla-
fully transmitted to the beet price, and vice tion? We treat the introduction of Vermont’s
versa. The prices were perfectly co- GE labeling law as a quasi-natural experi-
integrated. However, following the July 2016 ment making use of two findings from the
break, the correlation between the two price previous section to identify impact. First,
series fell to 0.57, indicating a breakdown in U.S. sugar prices were highly correlated with
the relationship. world sugar prices prior to the law taking ef-
Within the U.S. market, the July 2016 fect (rows 3 and 4 of table 4). Second, world
break appears to have been isolated to re- prices were unaffected by the Vermont law
fined sugar. Turning to row 2 of table 4, the (columns 7 and 8 of table 4). Taken together,
correlation between the U.S. refined cane these factors suggest that world sugar prices
price and the U.S. raw cane price fell from constitute an almost-ideal control against
0.96 to -0.12 following the break. The correla- which to assess the effects of the Vermont
tion between U.S. refined beet and U.S. raw law.
cane went from being highly positive (0.95) One method by which to identify the treat-
to highly negative (-0.74). These pre- and ment effects of a mandatory labeling law
Carter and Schaefer Impacts of Mandatory GE Food Labeling: A Quasi-Natural Experiment 9
introduction of the law (i.e., d ¼ 0; c ¼ b). structural break using the pre- and post-
Similarly, for each equation, we reject with structural-break coefficient estimates from ta-
99% confidence the hypothesis that the U.S.- ble 5. The counterfactual estimate (CF) for
world cointegrating relationship was the what U.S. refined sugar prices would have
same prior to and following the introduction been had the Vermont Law not taken effect
of the Vermont Law (i.e., d ¼ b). ^ wt . Post-regime
is calculated as CFt ¼ ^a þ b
These findings are entirely consistent with beet and cane prices are calculated as
the theoretical effects of mandatory GE label- ^ t ¼ ^a þ ^d þ ^c wt .10
P
ing. First, the law induces a price premium (or Constructed CF and Post-law-regime price
discount) for non-GE (GE) products relative to series are shown in logs in figure 4 and sum-
what the price would have been in the absence marized in levels in table 7. Our estimation
of the legislation. Second, GE and non-GE suggests that, in the absence of GE labeling
products are no longer interchangeable in legislation, the average U.S. refined sugar
regions that are de facto subjects to labeling price would have been approximately
requirements. Thus, such laws reduce integra-
tion with external markets that continue to treat
the GE and non-GE products as homogenous. 10
Note that we have actual observations for “post-regime”
The total impact of the Vermont labeling beet and cane prices. We use our predicted estimates instead of
law on U.S. refined sugar prices is the aggre- actual prices to isolate the impact of Vermont law. A variety of
gate effect from the two forces discussed demand and supply factors not related to the Vermont law can
impact U.S. and world prices in any given month. By focusing on
above. We estimate this impact for the 12- the average relationship over time, we eliminate (or at least sub-
month period immediately following the stantially reduce) the biases created by these extraneous factors.
Carter and Schaefer Impacts of Mandatory GE Food Labeling: A Quasi-Natural Experiment 11
Table 7. Impact of GE food Labeling Laws and other import restrictions for international
on U.S. Sugar Prices suppliers. These supply constraints drive U.S.
prices above world prices, and—even in light
Table 8. U.S. Sugar Industry Gross Receipts (GE Law vs. No Law Scenario)
Gross Receipts, July ‘16–June ‘17
Deliveries GE Law No Law Difference
(1,000 STRV) (Million Dollars)
Beet 5,188 $2,961.56 $3,396.97 $435.40
Cane 6,187 $4,091.25 $4,050.77 $40.48
Note: Gross receipts are obtained by multiplying GE Law & No Law prices by domestic deliveries (1.07 raw-to-refined conversion rate).
The substantial imbalance between the the deadweight loss of GE labeling require-
losses to beet processors and the gains to ments. Losses to beet processors are in part
cane refiners is only partically connected to offset by a gain to food manufacturers
Carter and Schaefer Impacts of Mandatory GE Food Labeling: A Quasi-Natural Experiment 13
purchasing beet sugar. Likewise, the gain to are almost perfectly correlated over the sam-
cane refiners is offset in part by a loss to food ple period (table 4).
manufacturers purchasing cane sugar. Lower Next, we consider confounding factors.
costs to GE products may also be passed Beginning in January 2011, Mexican sugar
through to consumers in the form of exports to the U.S. increased substantially.
lower food prices. The magnitude of these This surge in imports gave rise to antidump-
offsets—and in turn the deadweight loss—is ing and countervailing duty (ADCVD) pro-
dependent on the elasticities of demand in ceedings, which, in December 2015,
each market. culminated in volumetric and price restric-
tions on Mexican sugar exports to the United
States (USITC 2015). Carter, Saitone,
and Schaefer (2017) find that these restric-
Robustness and Sensitivity
tions impacted the U.S.-world price relation-
ship. We control for this by including an
In this section, we use alternative specifica- indicator variable in the regime-switching
tions to examine the robustness of our model. model equal to unity for all time periods after
We consider the inclusion of alternative and December 2015 (and zero otherwise). The
additional controls, the possible existence of world price is again the world refined price.
confounding factors, and the implications of Estimation results for this specification are
our time-series data. The percentage impact reported in columns (3) and (4) of table 9.
of Vermont labeling requirements on U.S. re- Coefficient estimates imply the Vermont law
fined sugar prices implied by these alternative reduced beet prices by 16% and increased
specifications ranges from 1% to 13% for cane prices by 13%.
cane and 10% to 16% for beet. Another potential estimation issue relates
Columns (1) and (2) of table 9 present the to the time-series properties of our data.
results from re-estimating the regime- Spurious correlation caused by non-
switching model shown in equation 2 using stationarity in our data could lead to incor-
the world raw price as the control variable rect inference. The risk of spurious correla-
rather than the world refined price. tion is low in this setting because of the
Consistent with the findings in table 5, the commodity nature of sugar—market prices
Vermont law drives a wedge between U.S. move together. However, for the sake of ro-
and world prices and reduces the level of in- bustness, we correct for non-stationarity via
tegration between the two markets. These first-difference estimation.
results imply a price impact that is nearly Results from re-estimating equation (2) in
identical to that discussed above for both first differences are reported in columns (1)
cane and beet prices. This is not surprising and (2) of table 10, respectively, for beet and
because world raw and refined sugar prices cane prices. First-differencing substantially
14 August 2018 Amer. J. Agr. Econ.
reduces the precision of our estimates, but harsh penalties for food manufacturers found
findings are unchanged in substance. Turning to be in violation of labeling requirements.
first to column (1), the GE law drives a 2% Congress responded soon after the Vermont
(positive) wedge between U.S. beet and law took effect with the successful promulga-
world refined prices. Coefficients on first- tion of the National Bioengineered Food
differenced world prices are insignificant. Disclosure Standard (NBFDS). The NBFDS
The point estimate falls from 0.015 to 0.003. nullifies all state-level attempts to establish
Combining the two effects implies a 10% re- mandatory labeling rules, and instead
duction in U.S. refined beet prices as a result imposes disclosure requirements at the fede-
of the Vermont Law. Cane results in column ral level. Many aspects of the NBFDS are
(2) show a 1.7% increase in Cane prices as a currently subject to public comment and re-
result of the Labeling Law. Coefficients on main to be finalized.
world refined prices are significant at 99% In the period surrounding the implementa-
both before and after the imposition of the tion of Vermont Act 120, commodity markets
law, but the post-law coefficient is not signifi- responded. In this research, we investigate
cantly different from the pre-law coefficient. the response in the U.S. sugar market. In
In columns (3) and (4), we add the indicator mid-2016, refined cane sugar began selling at
variable that controls for the imposition of a substantial premium over refined beet
the U.S.-Mexico ADCVD suspension agree- sugar. Our analysis supports the explanation
ments. Results are similar to those in columns that the divergence in U.S. prices for refined
(1) and (2) and imply a 9.8% reduction in cane and beet sugar was the result of
beet prices and a small ( 1%) increase in Vermont’s mandatory GE labeling. The di-
cane prices as a result of the law. vergence occurred on or around July 2016—
the month the Vermont Act took effect.
Counterfactual price estimates generated
Conclusion by a regression model suggest that GE food
labeling initiatives generated a small pre-
In the United States, a push for mandatory mium for cane sugar and a price discount for
labeling of GE foods began in Oregon, beet sugar of approximately 13% relative to
California, and Washington and rippled what prices would have been in the absence
through at least one-half of all U.S. states. In of such legislation. An open question is
many ways, Vermont Act 120, which passed whether the new cane-to-beet price wedge is
in May 2014 and took effect in July 2016, was permanent or, alternatively, whether prices
the culmination of those efforts. The law re- will converge again once the U.S. Food and
quired that (with a few exemptions) foods Drug Administration (FDA) has offered final
containing GE ingredients sold for home con- guidance on NBFDS compliance. Increased
sumption be labeled. The law also established domestic deliveries by U.S. beet processors in
Carter and Schaefer Impacts of Mandatory GE Food Labeling: A Quasi-Natural Experiment 15
the face of falling beet prices suggests that Journal of the Econometric Society 61
growers believe the price wedge may be per- (4): 821–56.
manent. Implications extend beyond the U.S. Artuso, A. 2003. Risk Perceptions,
European Journal of Law and Economics Lusk, J.L., and A. Rozan. 2008. Public Policy
16 (3): 327–44. and Endogenous Beliefs: The Case of
Dickey, D.A., and W.A. Fuller. 1979. Genetically Modified Food. Journal of