Cartel Behavior. Why does collusion occur? In the competitive model, firms enter until the last firm earns zero economic profits

Transcription

1 Cartel Behavior Why does collusion occur? In the competitive model, firms enter until the last firm earns zero economic profits Competition is clearly tough on firms In economic theory, cartels form in order to increase profits above the competitive level. Quote: Organized crime in America takes in over forty billion dollars a year and spend very little on office supplies, Woody Allen.

2 Cartel Behavior A very simple (but useful!) first model of a cartel is that they behave as a monopolist An assumption in our competitive model was that firms were price takers A firm s revenue is pq That is, firms take prices as given to them by the market

3 Cartel Behavior A cartel, however, coordinates the decisions of firms By doing so, it can choose any price quantity pair on the demand curve

4 Cartel Price/Quantity Choices Price P 1 P 2 P 3 D Quantity Q 1 Q 2 Q 3

5 Cartel Behavior The total revenue to the cartel is: TR(Q)=P(Q)Q Where P(Q) is the inverse demand curve Let C(Q) be the total cost to the cartel for producing output of Q For simplicity, let s assume that all N firms in the cartel are identical Let s also assume that they allocate the market shares evenly, i.e. q=q/n Then C(Q)=Nc(q) where q=q/n

6 Cartel Behavior The cartel s total profit is: TR(Q)-C(Q) The profit maximizing rule for the cartel is to set marginal revenue equal to marginal cost Marginal revenue, en e MR(Q) is the additional revenue en e from producing one additional unit Claim: If the demand curve is linear, it can be shown that marginal revenue has twice the slope of the demand curve If MR(Q) >MC(Q) then profits can be increased by producing one more unit If MR(Q)<MC(Q) then profits can be increased by producing one less unit

7 Cartel Versus Competitive Outcomes P MC P* MR Market Demand Q=nq Q*

8 Cartel Behavior Price is greater than the competitive level Quantity is less than the competitive level l -The cartel uses its market power -Restricts the supply below the competitive level in order to increase profits

9 Cartel Behavior The marginal costs for the cartel is less than the price This generates a deadweight loss This deadweight d loss comes from not having competitively functioning markets (which are efficient and have no deadweight loss)

10 Inefficiency of Collusion P P* Deadweight Loss MC MR Market Demand Q=nq Q*

11 Cartel Behavior Cartel Members have an incentive to cheat mc(q)<p for all members of the cartel The marginal cost to a member is less than the price Cartel members are therefore tempted to expand output above q=q/n An extremely important t problem for a successful cartel is how to monitor and prevent such cheating

12 Empirical Regularities Hay and Kelly, Journal of Law and Economics, Examined Department of Justice memoranda on collusion investigations from Jan 1963-Dec Cases that generated guility verdicts or nolo contendere pleas. Horizontal cases (vertical cases excluded). Remark: These were the cartels that were caught.

13 Hay and Kelly Outline: 1. Factors that t facilitate t collusion. 2. The Data 3. Results.

14 Factors Which Facilitate Collusion 1. Fewness of numbers. The smaller the number of people in the conspiracy, the less likely that disagreements will occur. Easier to detect cheating.

15 Hay and Kelly 2. Product Homogeneity. If the product is not complicated, it is simpler to negotiate collusive agreements. If products are stable over time, conspiracies more likely to persist.

16 Hay and Kelly 3. Demand Inelasticity. The rewards to collusion are greater with inelastic demand. Need to reduce output by less. 4. Sealed Bidding. If all prices are publicly announced, it is easier to monitor cheating on the collusive arrangement. Highly transparent markets most susceptible to collusion

17 Hay and Kelly 5. Industry Social Structure. Ad dominant tfigure exists to lead dthe cartel.

18 Hay and Kelly The authors use fact memoranda and other supporting documents created by the DOJ. Most collusion was found by a complaint from a competitor, customer or a grand jury investigation. Table 2 shows: 1) In some cases, larger groups (more than 10) do conspire, but they usually have a trade assocation. 2) In 7/8 cases with 16 or more firms, there was a trade assocation.

19 Hay and Kelly Number of Conspirators and Trade Assn. Involvement Number of >25 Total Conspirators Number of Cases Trade Assn. Involvement Na 6 18

20 Hay and Kelly 3) A comparison of the number of conspirators with the number of firms shows that not all firms typically collude.

21 Hay and Kelly Most of the cases are in markets with homogenous products. The fact memoranda showed there were examples of dominant individuals who helped to lead the cartel.

22 Hay and Kelly Bid-rigging was a factor in 15 cases. Bidding is often for lumpy projects. More concentrated markets tended to have longer last conspiracies. i

23 Factors that Facilitate Cartel Formation Three factors are needed to make collusion successful: 1) Deter entry/increased competition by noncartel firms. 2) Expected punishment low compared to expected gains from collusion. 3) Cost of enforcing agreement must be low relative to expected gains.

24 Limiting Competition by Non- Cartel Firms. If demand is very inelastic, industry profits can be substantially increased by reducing quantity from the competitive level. However, this might increase entry by new firms. New to have effective barriers to entry such New to have effective barriers to entry such as high entry costs, control over strategic input, or intimidation of entrants.

25 Cost of Enforcing Agreements Rises with number of firms in the market. Rises as products become more differentiated. Rises as demand d is more volatile. Falls as transaction data becomes more available. Falls if an industry association exists.

26 Where Collusion is Most Prevalent The number of firms is small. The products are homogenous. Demand does not fluctuate a lot. The actions of other colluders can easily be observed.

27 Methods of Preventing Cheating Divide market by buyers or geographic areas. A most favored nation clause. Promise all customers that they will get the best price. If firms lower the price, they have to compensate past customers. Meeting the competition clauses. Agree to match the price of all competitors. Trigger price: If price falls below a certain level, a price war is anticipated.

28 Some Facts Hay and Kelly (1974) examine 62 cases of collusion Fact memoranda prepared by Department of Justice Antitrust Staff Section 1 (Sherman act) cases which were won in trial or nolo contendre pleas from Jan 1963 to Dec 1972 Data is a bit sketchy -only included cartels that are caught -set of explanatory variables a bit limited -cross industry study However, findings are representative of other later studies and conventional wisdom

29 Number of Conspirators Intuition and economic theory both suggest that collusive arrangements will be easier to agree upon when the number of conspirators is small Monitoring collusive arrangements is also easier when the number of conspirators is smaller Easier to detect price cuts or expansion of output

30 Number of Conspirators Number of Conspirators and Trade Assn. Involvement Number of >25 Total Conspirators Number of Cases Trade Assn. Involvement Na 6 18

31 Number of Conspirators The average number of conspirators is percent of the conspiracies involve ten or fewer firms In 7/8 cases with more than 15 firms, a trade industry association was involved Trade associations may facilitate collusion by improved market monitoring and cartel discipline Even without trade associations Many conspiracies were in fact highly organized with chairmen, rules of order, agendas and regularly scheduled meetings This is all consistent with our model which predicts that members have incentives to cheat Monitoring and discipline is central to a functioning cartel

32 Concentration Table 2 also displays 4 firm concentration ratio Higher concentration may help in monitoring and planning cartel agreements Low concentration, easy to enter industries may not be able to sustain collusive prices In 38/50 cases, concentration was above 50 percent

33 Homogeneity Hay and Kelly argue that product homogeneity is high for the markets in their data set With a homogenous product, the cartel has to negotiate fewer price/quantity schedules for the collusive arrangement Simplifies bargaining and monitoring

34 Market Transparency Many of the cartels involved sealed bidding for government contracts Sealed bidding simplifies the cartels monitoring problem Competitive bids are typically announced very shortly after the tender Deviations can be detected immediately More generally, highly transparent markets are more subject to collusion

35 Market Transparency Spectrum Auctions- government allocates rights to run mobile telephone licenses and other wireless services for a particular geographic g market Simultaneous ascending auction -many licenses for sale at once -bids increase over the course of the auction Small number of firms win most of the licenses Bidding occurs over 3 or more months Bids and identity of bidders are observed in real time

36 Market Transparency Crampton and Schwartz (2000) find brazen instances of collusion For example, firms use the license number in the trailing digits of bids to signal rivals Firms submit retaliatory bids in order to limit competition for most preferred license

37 Market Transparency The Federal Communications Commission disguised the identity of bidders in most recent auctions Bajari and Yeo (2009) find that retaliatory bids fall Also, a higher percentage of the bids are straightforward ag a That is, bidders appear to submit slowly increasing bids on most preferred items as the rounds of the auction progress As markets become less transparent, the ability of firms to collude is lessened

38 Price Wars A final prediction of our theory is that collusive markets may be subject to price wars This is particularly true if the market is subject to independent demand/cost shocks or the market is not perfectly transparent Porter (1983) studies Joint Executive Committee Cartel of US railroads that operated in the 1880 s prior to the Sherman act JEC controlled eastbound freight from Chicago in the 1880s.

39 Price Wars JEC office took weekly accounts so that the shipments could be monitored. Demand was quite variable and hard to predict Cheating on collusion, as reported by Railway Review, occurs 40 percent of the time. Price Wars had an average duration of about ten weeks Price was 66% higher in cooperative periods and quantity 33% lower. As a whole, the cartel could expect to earn 11% higher revenues during cooperative periods

40 Price Wars

41 Factors that Facilitate Cartel Formation Three factors are needed to make collusion successful: 1) Deter entry/increased competition by noncartel firms. 2) Expected punishment low compared to expected gains from collusion. 3) Cost of enforcing agreement must be low relative to expected gains.

42 Limiting Competition by Non- Cartel Firms. If demand is very inelastic, industry profits can be substantially increased by reducing quantity from the competitive level. However, this might increase entry by new firms. Need to have effective barriers to entry such Need to have effective barriers to entry such as high entry costs, control over strategic input, or intimidation of entrants.

43 Cost of Enforcing Agreements Rises with number of firms in the market. Rises as products become more differentiated. Rises as demand d is more volatile. Falls as transaction data becomes more available. Falls if an industry association exists.

44 Gains Versus Losses of Collusion Expected cost of collusion rises with probability of enforcement Gains are higher if demand is inelastic If l b i il bl i If many close substitutes are available, gains from collusion are lower

45 Methods of Preventing Cheating Divide market by buyers or geographic areas. A most favored nation clause. Promise all customers that they will get the best price. If firms lower the price, they have to compensate past customers. Meeting the competition clauses. Agree to match the price of all competitors. Trigger price: If price falls below a certain level, a price war is anticipated.

46 Bid Rigging Bid-rigging is one of the most commonly observed forms of collusion Economic models of competitive bidding have two robust predictions Prediction 1. Non-collusive bids should be independent d -Sealed bidding in non-collusive markets implies that a firm is ignorant of competitors bids -This implies that the bid should be uncorrelated (after controlling for observed cost information) -Collusive schemes typically induce correlation

47 Independent Bids 9 Plot of Noncollusive Bids 8 7 Firm 2 Bid Firm 1 Bid

48 Collusive Bids 10 Plot of Collusive Bids Firm 2 Bid Firm 1 Bid

49 Competitive Versus Collusive Bidding Three firms A,B and C Submit sealed bids for infrastructure project Low bidder wins Two cases: 1. A,B and C submit bids competitively 2. A and B collude, C submits a non-collusive bid How to distinguish collusive from competitive bids

50 Competitive Versus Collusive Bidding Firm Identity Cost for 1 st Competitive Collusive Bid Project Bid dfor 1 st for 1 st Project Project A $1.0 Million $1.19 Million $1.29 Million B $1.2 Million $1.2 Million Phony Bid>$1.29 Million C $1.3 Million $1.3 Million $1.3 Million

51 Competitive Versus Collusive Bidding Collusive Bid Firm Identity Cost for 2nd Competitive Project Bid dfor 2 nd for 2nd Project Project A $1.0 Million $1.19 Million $1.19 Million B $1.3 Million $1.3 Million Phony Bid > $1.19 Million C $1.2 Million $1.2 Million $1.2 Million

52 Competitive Versus Collusive Bidding Prediction 2. Bids reflect costs and the normal use of market power The competitive bids permute with the costs Th ll i bid d i h h The collusive bids do not permute with the costs

53 Bid Rigging Porter and Zona examine bids of convicted colluders in New York city paving and Ohio school milk. They find: 1. Colluders bids have a high, positive correlation 2. Non-winning bids are unrelated to costs -Closest firm typically wins -Second lowest bidder is not typically the second closest firm 3. Collusion raises bid levels compared to a price index of other markets

54 Bid Rigging Bajari and Ye examine construction bids in the Upper Midwest. Bids are well explained by cost controls and measures of market power -97 percent of the variation in bids can be explained by cost controls -closest firm is more likely to win -firm with the lowest backlog is more likely to win Bids are independent after controlling for costs

55 Collusion in Wisconsin DOT Vinton Construction and Streu Construction were fined for rigging g bids on Wisconsin DOT projects They also hired an employee at James J. Cape Co. to provide them with inside information about upcoming bids The bids are much more correlated with each other than competing bids

56 1.05 WI DOT Road Construction Bids, Vinton Construction vs. Streu Construction 1 Vinton's total bid 0.95 y = 0.393x (p=0.030, 0.000) R² = Streu's total bid. reg bidtotalv bidtotals, robust Linear regression Number of obs = 36 F( 1, 34) = 5.83 Prob > F = R-squared = Root MSE = Robust bidtotalv Coef. Std. Err. t P> t [95% Conf. Interval] bidtotals _cons reg bidtotalv bidtotals Source SS df MS Number of obs = F( 1, 34) = 6.79 Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = bidtotalv Coef. Std. Err. t P> t [95% Conf. Interval] bidtotals _cons

57 1.05 WI DOT Road Construction Bids, Vinton Construction vs. James Cape & Sons Vinton's total bid y = x (p= 0.022; 0.000) R² = James Cape's total bid. reg bidtotalv bidtotalj, robust Linear regression Number of obs = 35 F( 1, 33) = 3.47 Prob > F = R-squared = Root MSE = Robust bidtotalv Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalj _cons reg bidtotalv bidtotalj Source SS df MS Number of obs = F( 1, 33) = 4.78 Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = bidtotalv Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalj _cons

58 1.05 WI DOT Raod Construction Bids, Vinton Construction vs. Zignego Companies 1 Vinton's total bid y = 0.167x (p=0.653; 0.000) R² = Zignego's total bid. reg bidtotalv bidtotalz, robust Linear regression Number of obs = 27 F( 1, 25) = 0.30 Prob > F = R-squared = Root MSE = Robust bidtotalv Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalz _cons reg bidtotalv bidtotalz Source SS df MS Number of obs = F( 1, 25) = 0.56 Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = bidtotalv Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalz _cons

59 1.15 WI DOT Road Construction Bids, Streu Construction vs. James Cape & Sons 1.1 Streu's total bid y = x (p=0.004; 0.000) R² = James Cape's total bid. reg bidtotals bidtotalj, robust Linear regression Number of obs = 35 F( 1, 33) = 4.53 Prob > F = R-squared = Root MSE = Robust bidtotals Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalj _cons reg bidtotals bidtotalj Source SS df MS Number of obs = F( 1, 33) = 3.72 Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = bidtotals Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalj _cons

60 1.04 WI DOT Road Construction Bids, Streu Construction vs. Zignego Companies 1.02 Streu's total bid y = x p=(0.533; 0.000)) R² = Zignego's total bid. reg bidtotals bidtotalz, robust Linear regression Number of obs = 26 F( 1, 24) = 1.41 Prob > F = R-squared = Root MSE = Robust bidtotals Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalz _cons reg bidtotals bidtotalz Source SS df MS Number of obs = F( 1, 24) = 0.87 Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = bidtotals Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalz _cons

61 WI DOT Road Construction Bids, Zignego Companies vs. James Cape & Sons Zignego's total bid y = x (p=0.768; 0.049) R² = James Cape's total bid. reg bidtotalz bidtotalj, robust Linear regression Number of obs = 27 F( 1, 25) = 0.40 Prob > F = R-squared = Root MSE = Robust bidtotalz Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalj _cons reg bidtotalz bidtotalj Source SS df MS Number of obs = F( 1, 25) = 1.09 Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = bidtotalz Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalj _cons

62 1.4 WI DOT Road Construction Bids, Lunda Construction vs. Edward Kraemer & Sons Inc. 1.3 Lunda's total bid y = x (p=0.383; 0.002)) R² = Edward's total bid. reg bidtotall bidtotale, robust Linear regression Number of obs = 619 F( 1, 617) = 0.20 Prob > F = R-squared = Root MSE = Robust bidtotall Coef. Std. Err. t P> t [95% Conf. Interval] bidtotale _cons reg bidtotall bidtotale Source SS df MS Number of obs = F( 1, 617) = 4.10 Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = bidtotall Coef. Std. Err. t P> t [95% Conf. Interval] bidtotale _cons

63 WI DOT Road Construction Bids, Zenith Tech. Inc. vs. Edward Kraemer & Sons Inc. 1.3 Zenith's total bid y = x (p=0.228;0.008) R² = Edward's total bid. reg bidtotalz bidtotale, robust Linear regression Number of obs = 257 F( 1, 255) = 2.10 Prob > F = R-squared = Root MSE = Robust bidtotalz Coef. Std. Err. t P> t [95% Conf. Interval] bidtotale _cons reg bidtotalz bidtotale Source SS df MS Number of obs = F( 1, 255) = Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = bidtotalz Coef. Std. Err. t P> t [95% Conf. Interval] bidtotale _cons

64 1.5 WI DOT Road Construction Bids, Zenith Tech. Inc. vs. Lunda Construction 1.4 Zenith's total bid y = x (p=0.900;0.008) R² = Lunda's total bid. reg bidtotalz bidtotall, robust Linear regression Number of obs = 482 F( 1, 480) = 2.48 Prob > F = R-squared = Root MSE = Robust bidtotalz Coef. Std. Err. t P> t [95% Conf. Interval] bidtotall _cons reg bidtotalz bidtotall Source SS df MS Number of obs = F( 1, 480) = Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = bidtotalz Coef. Std. Err. t P> t [95% Conf. Interval] bidtotall _cons

65 Documentation of Preliminary Results from Wisconsin, Minnesota, and Illinois Departments of Transportation (DOT) Datasets of Highway Procurement Bids This document contains regression results and histograms of Wisconsin, Minnesota, and Illinois bids. These states are chosen to contrast the markedly different bid patterns between a known case of price collusion in Wisconsin and non-collusive firms in the other two states. Non-collusive firms from Wisconsin are also included to contrast with the collusive ones in the same state for robustness. Our result has never been documented to the best of knowledge. Total bids, as opposed to unit bids, are used in this preliminary analysis. A total bid is defined as the sum of unit bids of all items involved multiplied by quantity. Since projects are of wide range of sizes, all total bids are normalized, by dividing by engineer s estimates (or average, if estimate is not available). These three datasets are constructed from bid abstracts publicly available on the respectively DOT websites. The results are divided into four sections: (I) Collusion Case in Wisconsin; (II) Non-Collusion Cases in Wisconsin; (III) Non-Collusion Cases in Minnesota; and (IV) Cases in Illinois. (I) Collusion Cases in Wisconsin In November 2005, the owners of Vinton Construction Co. (Two Rivers, WI) and Streu Construction Co. (Manitowoc, WI), together with a former employee of James Cape & Sons Company (Racine, WI) were convicted of bid-rigging in highway procurement auctions, involving illegal exchange of price information and private allocation of projects among the three firms. 1 More than 30 projects were involved from 1997 to Our first approach is to regress James Cape & Sons bids on Streu and Vinton s bids, among the contracts where all three firms submitted bids. This is the regression result with robust standard errors:. reg bidtotalj bidtotalstreu bidtotalvinton, robust Linear regression Number of obs = 35 F( 2, 32) = 7.37 Prob > F = R-squared = Root MSE = Robust bidtotalj Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalst~u bidtotalvi~n _cons News release from the Office of Inspector General can be found at 1

66 Both Streu and Vinton s coefficients are negative and statistically significant. This reflects that the three firms indeed allocated projects among themselves: once the winner was internally decided, it would submit a low bid while the other two firms would submit significantly higher bids, as pretense of competition. Our second approach looks at the distribution of difference in bids for pairs of firms among the three colluders: 2

67 Two observations are that (i) these distributions are not normal or bell-shaped, and that (ii) these distributions are not centered at zero. 3

68 (II) Non-Collusion Cases in Wisconsin Edward Kraemer & Sons Inc., Lunda Construction Co., and Zenith Tech. Inc. were chosen as the three firms among non-colluders that bid in the most number of contracts. We first regress Edward s bids on Lunda and Zenith s bids:. reg bidtotale bidtotall bidtotalz, robust Linear regression Number of obs = 253 F( 2, 250) = 0.99 Prob > F = R-squared = Root MSE = Robust bidtotale Coef. Std. Err. t P> t [95% Conf. Interval] bidtotall bidtotalz _cons The coefficients, although still negative, are much smaller in magnitude and are, in fact, not statistically different from zero. Also note a much smaller R 2 value than the regression in the previous section: a fellow non-collusive competitor s bids have much less explanatory power than a fellow colluder s. Now we plot the bid difference between pairs formed among these three non-collusive firms: 4

69 5 These distributions stand in stark contrast with those in the previous section that (i) they are roughly normal or bell-shaped, and that (ii) they are centered at zero. This indicates that (i) difference between bids among non-colluders are not premeditated, and furthermore, that (ii) profit margins are tight and roughly the same among bidders under competition.

70 (III) Non-Collusion Cases in Minnesota We repeat the same exercise for three pairs of Minnesota firms, chosen because they have the largest number of contracts where both firms bid in. Note the statistically significant positive regression coefficients as well as the roughly normal or bell-shaped distribution of bid differences, a feature shared by non-collusive firms in Wisconsin. (1) Between Bauerly Bros Inc. and Duininck Brothers Inc.:. reg bidtotalnormb bidtotalnormd, robust Linear regression Number of obs = 86 F( 1, 84) = Prob > F = R-squared = Root MSE = Robust bidtotalno~b Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalno~d _cons

71 (2) Between Bauerly Bros Inc. and Central Specialties Inc.:. reg bidtotalnormb bidtotalnormc, robust Linear regression Number of obs = 83 F( 1, 81) = Prob > F = R-squared = Root MSE = Robust bidtotalno~b Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalno~c _cons

72 (3) Between Hardrives Inc. and Valley Paving Inc.:. reg bidtotalnormh bidtotalnormv, robust Linear regression Number of obs = 63 F( 1, 61) = Prob > F = R-squared = Root MSE = Robust bidtotalno~h Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalno~v _cons

73 (IV) Cases in Illinois We again repeat the same exercise for four pairs of Illinois firms, chosen because they have the largest number of contracts where both firms bid in. Note that regression coefficients are rarely statistically different from zero and that distribution in bid difference is not normal or bell-shaped. (1) Between K-Five Construction Co. and Gallagher Asphalt Co.:. reg bidtotalnormf bidtotalnorml, robust Linear regression Number of obs = 42 F( 1, 40) = 2.00 Prob > F = R-squared = Root MSE = Robust bidtotalno~f Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalno~l _cons

74 (2) Between Civil Construction Inc. and Brandt Construction Co.:. reg bidtotalnormc bidtotalnormr, robust Linear regression Number of obs = 33 F( 1, 31) = 3.68 Prob > F = R-squared = Root MSE = Robust bidtotalno~c Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalno~r _cons

75 (3) Between E. T. Simonds Construction Co. and Perry County Construction Co.:. reg bidtotalnorme bidtotalnormp, robust Linear regression Number of obs = 31 F( 1, 29) = 0.02 Prob > F = R-squared = Root MSE = Robust bidtotalno~e Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalno~p _cons

76 (4) Between D. Construction Inc. and Gallagher Asphalt Co.:. reg bidtotalnormd bidtotalnorml, robust Linear regression Number of obs = 31 F( 1, 29) = 0.00 Prob > F = R-squared = Root MSE =.1022 Robust bidtotalno~d Coef. Std. Err. t P> t [95% Conf. Interval] bidtotalno~l _cons