# Brand-name investment of candidates and district homogeneity: an ordinal response model.

I. Introduction

Advertising is often viewed as a sunk investment that will receive a return only if firms do not cheat their customers on quality, much like a security bond that will be lost for non-performance of a contract. In the last period a firm will cheat if there is no future return from the investment [15]. Knowing that a firm is in it last period will then reduce sales in the last period to zero, so the firms will begin to cheat the period before the last. This is the usual iterated prisoner's dilemma (PD) game with a finite life; cheating begins in the first period and continues. Demsetz [10] argues that since firms are transferable through voluntary trade, the possibly infinite life of a firm transforms this finite, iterated PD game with a known end into an iterated PD game with no known end. As Axelrod [1] suggests, the best of the known strategies for this game is the tit-for-tat strategy. In this game, firms live up to quality claims if customers continue to pay for their product. Demsetz also argues that brand name cannot serve as a barrier to entry for alienable firms because the firm, with its brand name intact, can be purchased by buyers who are more efficient producers than the current producer.

Since professionals and politicians who have made investments in brand name have namebrand capital that is inalienable, these markets may suffer from the last-period problem [12; 17; 18; 19]. If the seller knows the last period because he intends to quit, cheating cannot be prevented by investment in brand-name capital. Still, uncertain ending periods may be enough to insure that the seller will produce at the advertised quality if buyers determine the final period [29]. V

Lott [18] has incorporated brand-name entry barriers into a model analyzing competition for legislative seats. Though Lott fundamentally accepts Demsetz's [10] critique of entry barrier analysis, Lott points out that Demsetz's reasoning does not hold in cases where the producer is not an ongoing enterprise, where the producer is a single individual, not a firm. Political brand-name investment erects a barrier to entry because this capital is not transferable.

Nelson [24; 25] distinguishes between the "search" and "experience" qualities in a good. Search qualities are those that can be discovered in the search process before purchase, and experience qualities are those that can be discovered only after purchase, as the product is used or consumed.

Nelson [26] argues that the repeat-purchase mechanism gives candidates some incentive to be truthful in their advertising. Further, the candidate has an incentive to direct his advertisement to those who would most likely vote for him, if they knew his position. Telser [29] and Ferguson [11] suggest that candidates are experience goods and so the voter knows that he cannot trust the politician to be completely honest and forthcoming. The voter instead relies on experience with that candidate (brand) or relies on trustworthy judges of candidates that act as agents for voters, i.e., the political party allowing its brand name to be used by the candidate.

The incumbent's advantage stems partially from his fund-raising advantage, which can be translated into advertising that builds brand-name capital. This fund-raising advantage cannot be translated into an advertising advantage for a candidate in a politically diverse district if the candidate is a search good [8; 12; 29]. These candidates' positions are well known and a greater amount of funds raised is viewed as a greater debt owed by the candidate to some supporting group: a group which some will oppose in diverse districts. Advertising becomes a signal about the level of debt owed to some interest group, which signals a worse buy; not a better buy, as is true if the candidate were an experience good. In homogeneous districts, investment in brand V name, by making the candidate's positions known to the constituency increases his expected vote margin and decreases the chance that a rival will come forward [8; 12; 29]. On the other hand, such investment in heterogeneous districts increases both the support and the opposition to the candidate.

Since the difference between search and experience goods depends on the cost of measuring the quality of a good before and after purchase, this cost difference can be changed by new technology and new institutions. Crain and Goff [8] examine the televising of legislatures, which is found in many, but not all, states. Crain and Goff postulate that politicians will be search goods in the states where legislatures are widely televised but will experience goods where they are seldom televised. They test their hypothesis using 1976 state house and senate races by stratifying their sample into high television exposure states and low television exposure states. Crain and Goff use population per seat as a measure of political diversity, reasoning that larger populations will be more diverse. To test their hypothesis, they regress population per seat, the number of multi-member districts, and seats up for election against number of winning challengers. They find that as population per seat increases, challengers win more often in high exposure states than in low exposure states. They interpret this result as support for the hypothesis that politicians are experience goods where the cost of information is high and search goods where the information costs are low.

Fremling and Lott [12] argue that politicians are not experience goods since investment in brand name will not prevent cheating due to the existence of the last-period problem. This prevents conspicuous investment in brand name from serving as a guarantee that the politician will not act contrary to his public political promises. They argue instead that politicians are search goods because voters learn the politician's utility function before any expected last period.

Brand name is not the only factor that may discourage challengers from coming forward in an election. More efficient production by a provider may keep potential entrants our of a market. This type of advantage and barrier is welfare enhancing. One example of this type of barrier is the advantage of tenure of the candidate that is not due to his sunk investment but to his investment in human capital through experience on the job. Other advantages, which we associate with brand name, may be due less to the candidate's efficiency and more to his sunk investment in goodwill.

In this paper we empirically explore the brand-name barriers to entry in British Parliamentary contests. Political entry barriers have previously been examined [6; 7; 16; 30]. We examine the effect on competition of important measures of brand-name investment identified elsewhere [3] and include measures of political diversity. Our measure of competitiveness is the excess of the number of candidates over the number of seats available. In the next section, we examine the consequences of information-cost differentials between British rural districts and urban districts in the last century. The third section discusses political diversity and brand-name advantage. Then, the data set is described, and the variables used in the empirical section are discussed. The fifth section presents a discussion of ordinal probit and ordinal logit models used to analyze the data and test our hypothesis. The results of the ordinal response models are discussed in the sixth section, including a test between the ordinal probit and ordinal logit specifications for our model.

II. The Cost of Standing and the Cost of Information

In Britain from 1832, the average cost of standing for a county seat was roughly 20,000 pounds in 1980 pounds, while a borough seat cost only about one-half that much [27]. Candidates paid more to run for county seats than for borough seats, though there were no residency requirements and county Members of Parliament (M.P.'s) lacked greater legislative powers than the borough M.P.'s. Also, neither the county M.P. nor the borough M.P. was paid a salary.(1) Why did candidates consistently pay twice the price of a borough seat to present a county district? One explanation is that the county seat was more secure than the borough seat [2]. The rural seat was more valuable because it was easier to retain once won.

Greater political security in the county seats can be explained by the higher cost of information in the rural districts. In the more densely populated urban districts, there was more V interaction with a greater variety of people, information spread by word of mouth more quickly and at lower cost. Following Crain and Goff [8], politicians are expected to be experience goods in countries and search goods in boroughs. Being an incumbent acts as a signal, a low-cost information source about the candidate. Investment in political brand name as campaign expenditure acts as a barrier to entry for the high information-cost county seats (though the real barrier to entry is the high-cost information). In the borough districts, lower information costs reduce incumbent advantage and the advantage from previous investment in brand name.

Telser [29], Nelson [26], and Crain and Goff [8] suggest that the degree of political diversity of the electorate influences the advantage of political brand-name investment. When politicians are search goods, as we expect them to be in boroughs, investment in brand name will be of little value in more diverse districts since politicians' positions are easily discovered without such investment and such investment suggests a debt owed to some special interest group. When politicians are experience goods, as we except them to be in counties, increasing political diversity will not decrease the value of brand-name investment since brand name informs the voters about otherwise unknown positions.

Different measures for political diversity can be used. Crain and Goff [8] use population per seat. A weakness of this measure is its questionable assumption that districts of the same size have the same degree of diversity, or that diversity increase with size. We propose an alternate measure, party homogeneity, which measures the diversity in part vote shares in recent elections in a districts. We use both Crain and Goff's measure and our own in testing the importance of brand-name investment in more versus less diverse districts and higher versus lower information-cost districts.

The hypothesis that we text requires some explanation. Since investment in brand-name capital in more homogeneous districts gives the candidate greater advantage than the same investment in a more diverse district, the interaction of brand-name investment variables with some measure of diversity should be negative (or positive with a measure of homogeneity). Furthermore, the slope of entry difficulty with respect to this interaction of diversity and past investment in brand name should be more negative in the low information-cost boroughs than the high information-cost countries. Information costs are hypothesized to be greater in the county districts than in the borough districts, leading us to expect greater importance in brand-name investment in reducing competition in the county districts. Investment in non-transferable brand-name capital, as

V measured both by past campaign victories and past campaign losses, should be more important in reducing the number of competitors in the county districts than in the borough districts. Our null hypothesis, then, is that the coefficient of the interaction of district diversity and past investment in brand-name will be no less in the lower information-cost boroughs than that coefficient for the county districts. Once we control for diversity with either measure (included separately in the regression), we can examine these interactions.

IV. The Data Set

Just as institutional and technological differences can lead to differences in the cost of pre-purchase information concerning candidate quality, sociological differences also can lead to such information-cost differentials. The difference explored in this paper is the difference between

V urban districts and rural districts before the coming of television and radio.

Crain and Goff [8] stratify their sample of states by high and low television exposure of the legislature, which effectively stratifies by cost of pre-purchase information about candidate quality. We stratify our sample by cost of information in another way, that is, by urban versus county districts. We postulate that in nineteenth century Britain, political information was transmitted either by talk or by print. Since it is less costly to talk with others and to read newspapers and pamphlets, we conclude that information costs are lower in the urban or borough districts than in the rural or county districts.

Since our original question is about the differential cost of standing in urban and rural districts in nineteenth century Britain, we have chosen British Parliamentary general elections from 1852 to 1880 as the population from which to draw our sample. Craig [4] has compiled results from all 381 districts in England, Wales, Scotland, and the Universities, leaving out the Irish constituencies from 1832 to 1885. We begin our data set with 1852 so that we can compile at least a twenty-year history of the candidates in that district before the first election in our data set. Further, we have taken a stratified sample of roughly sixty percent of English boroughs and counties, Scottish boroughs and countries, and Welsh boroughs and countries that were enfranchised in 1832. We have left out the five University districts because only two of these were enfranchised in 1832.

There are several features of this data set that make it especially useful for our purposes. First, it includes one easy-to-measure brand-name factor that can influence British Parliamentary elections, title of nobility. We also obtain a proxy for family electoral experience in that district because the data set includes full names and vote totals for all candidates during this period. Finally, the data set contains multi-seat districts with anywhere from one to three seats.

Economic theory predicts that it is more desirable for a candidate to contest an election (and therefore, there will be more candidates) when the rewards of office are higher, when the probability of winning is higher, or when the costs of challenging are lower. Our data set does not allow us to examine directly the issue of returns to office holding. Still, we can examine the strategies of individuals faced with an uncertain election result. Potential candidates generally are expected to have poor information regarding other potential non-incumbent candidates' chances of winning, and largely discount the effect of these candidates. We except that potential candidates are likely to focus on the current incumbents who are seeking re-election. The problem facing potential candidates is to decide whether to complete by judging their chances against the chances of incumbent-candidates.

A candidate's initial brand-name capital, the stock of information about the candidate in the hands of the voters in that district before the election gets under way, increases that candidate's chance of winning. If that candidate is an incumbent, the initial stock of brand-name capital also reduces the chances of other candidates and so should reduce the number of candidates opposing the incumbent. If the incumbent has won several elections in the past, has the family name of a former M.P. from the district, has run in several past elections albeit unsuccessfully, or has a title of nobility, the information about the incumbent candidate is greater at the beginning of the campaign, so initial brand-name capital of that incumbent candidate is greater. The main variables in the empirical model measure initial brand-name capital.

The observational unit for this study is the district election, not the seat. The data set contains variables by district and year for 200 districts

V and the election years from 1852-80. The variables are:

1. LOSERS: the dependent variable. This variable is the number of candidates in excess of the number of seats available, or the number of losing candidates. The values for LOSERS range from zero to four. An election is considered to be more competitively fought by the extent the number of candidates exceeds the number of seats.

2. VICTORIES: the total number of elections incumbent-candidates have won in this district. Previous successful campaign should increase name recognition and one's electoral strength in subsequent elections. Still, tenure also increases experience, and so, the efficiency in income redistribution activities.

3. DEFEATS: the total number of times incumbent-candidates have run unsuccessfully in this district. Such office seeking may contribute to name recognition, increasing one's chances in subsequent elections. Whereas VICTORIES measures both investment in brand-name capital and efficiency, this variable measures only investment in brand-name capital.

4. TITLED: the number of incumbent-candidates who are titled. Since title has brand name connotations, this variable should be positively related to LOSERS. Lott [17] states that brand name acquired outside of political markets may be valuable in a political market.

5. FAMILY: the number of incumbent-candidates who have the same last name as previous winning candidates in the district. Laband and Lentz [16] and Lott [17] suggest that while brand name cannot be traded, it can be handed down. This is way family name is considered important in this study.

6. BOROUGH: a dummy variable with a value of one if the election is in a borough and zero V if it is in a county.

7. BOUNDARY: a dummy variable to capture the effect of the 1868 Parliamentary Boundary Act. This Act reapportioned the legislative districts, adding new districts, combining districts and eliminating some districts [4]. New districts added by the Act were excluded from our sample because we need the history of participation from 1832-52 to create several independent variables. We include the dummy to capture the effect of this redistricting on competition. We could not control for other boundary changes that may have occurred occasionally throughout the period

V of study.

8. WALES and ENGLAND: two dummy variables to account for possible differences in competitiveness in the three countries.

9. FIFTY-SEVEN, FIFTY-NINE, SIXTY-FIVE, SIXTY-EIGHT, SEVENTY-FOUR and EIGHTY: dummy variables for the election years. 1857, 1859, 1865, 1868, 1874, and 1880, respectively. These variables are used to capture differences among general elections (year of the general election).

The next two independent variables, party homogeneity (HOMOGENEITY) and electorate size per seat (SIZE), measure the homogeneity or the lack it, of the district. Crain and Goff's measure assumes that larger electorates are less homogeneous than smaller districts [8]. We believe, however, the Crain and Goff's measure of homogeneity is unsatisfactory since in particular data sets, such as ours, certain large districts may be more homogeneous than smaller districts. Our alternate measure, HOMOGENEITY, is calculated as: (1) [Mathematical Expression Omitted] where LIBERAL [VOTE.sub.t] is the total votes for liberal candidates in election r, and TOTAL [VOTE.sub.t] is the total votes in election t. The variable HOMOGENEITY will range from zero to fifty, such that HOMOGENEITY plus fifty is the average percent of the votes cast over the past four elections for the predominant party.

Using this data set in previous study, Coats and Dalton [3] found that the variables VICTORIES, DEFEATS, TITLED, and FAMILY were correlated with whether and election was contested. If a district were not contested, the barriers to entry were too high for challengers to overcome, but if an election were held, any entry barriers were insufficient to forestall competitors. We believe that LOSERS measures the degree of competition in an election better than a dichotomous variable that is based on contested versus uncontested elections. As the number of candidates increased, holding the number of seats constant, entry barriers in the district fell as hurdle for challengers. Therefore, LOSERS is a measure of the size of the barriers to entry, or the degree of competition for a seat.

V. The Empirical Models

We estimate the following model, using two different measures of homogeneity, HOMOGENEITY and SIZE, and using two different econometric specifications, ordinal probit and ordinal logit: (2) LOSERS = f (FAMILY, TITLED, VICTORIES, DEFEATS, ENGLAND,

where [H.sup.a] is HOMOGENEITY in one case and SIZE in the other.

The homogeneity (or heterogeneity) variables are included in the models alone and in interaction terms with VICTORIES and DEFEATS (measures of past investment in brand name capital and signals of electoral strength). The theory suggests that the more homogeneous the district, the greater the impact of brand-name investment on reducing competition, since brand-name signals a better buy for the voters in these districts. In higher information-cost districts, such as our county districts and Crain and Goff's [8] low televised legislature states, this brand name signal should be more important. We include homogeneity by itself since we except that districts with a more even split between party allegiances of voters will be more fertile ground for opposition parties to make inroads and should be more highly contested districts.

The variables, DEFEATS and VICTORIES, provide measures of previous investment in political brand name and are used to examine Lott's [17] brand-name hypothesis. Though VICTORIES measures both efficiency in arranging transfer and previous investment in brand-name capital, DEFEATS measures only previous investment in brand-name capital. The variable BOROUGH and the BOROUGH dummy-interaction variables, BOROUGH*VICTORIES, and BOROUGH*DEFEATS are used to examine the hypothesis of differential election cost developed in this paper. The other variables are included to control for other effects over the various districts and election years.

The major consideration in the econometric estimation is the special feature of the data. In particular, the sample space of the dependent variable is the set of nonnegative integers. Here, standard econometric techniques that do not consider this special feature of the data, such as ordinary least-squares regression, present well-known problems [14; 20].

The recent developments in the analysis of count data have contributed significantly in this aspect of econometric modeling. One approach to the analysis of count data is through the estimation of nonlinear regression [14; 22]. But, these models do not consider the discrete and nonnegative nature of the count data. Another approach is through the specification of certain probability distributions such as Poisson distribution. The other approach, adopted in this study, is to treat the dependent variable as ordinal. To this end, McKelvey and Zavoina [23] have developed one form categorical model, known as the response model.

In the following we present the ordinal response model as it appears in the literature. Denote the response function as (3) [Mathematical Expression Omitted] where for observation i, [Mathematical Expression Omitted] is the theoretical response variable of interest, [x.sub.i] is a vector of explanatory variables, [Beta] is a vector of unknown parameters, and [u.sub.i] is a random disturbance term with E([u.sub.i/x.sub.i]) = 0 for all i. Then, [Mathematical Expression Omitted] for all i. Although [Mathematical Expression Omitted] is not observed, it belongs to a known category associated with an integer value for the observed variable [y.sub.i]. Following McKelvey and Zavoina [23], define a vector set of m + 1 constants [Mu] = ([[Mu].sub.o], [[Mu].sub.1,...,[[Mu.]sub.m]), such that [Mathematical Expression Omitted], where m is the number of categories. For the ith observation, [Mathematical Expression Omitted] belongs to the jth category if and only if (4) [Mathematical Expression Omitted] Also, define a dichotomous variable [Z.sub.ij] = 1 if [Mathematical Expression Omitted] falls in the jth category; = 0 otherwise. Then, for an observation i, condition (4) translates into the following probability in reference to (3): (5) [Mathematical Expression Omitted] where F is a cumulative distribution function. For a sample of size n, the likelihood function is (6) [Mathematical Expression Omitted]

If the distribution function F is specified as the cumulative standard normal function, equation (6) is the likelihood function for the ordinal probit model developed in McKelvey and Zavoina [23] and presented in Maddala [20, 46-49]. If F is specified as the logistic function, then (6) is the likelihood function for the ordinal logit model. In either case, the maximum likelihood (ML) method can be applied in the estimation of the slope coefficients [Beta] and the ancillary parameters [Mu]. Pratt [28] has shown that the log-likelihood functions for the ordinal probit model and the ordinal logit model are globally concave. Therefore, for both models the ML estimates are unique. For statistical inferences, like most other ML estimation, inverting the information matrix yields the variance-covariance matrix of the ML estimates that can be used for tests of significance [13]. To find the significance of all or a subset of coefficients, likelihood-ratio tests can be performed based on the log-likelihood function values.(2) In addition, an estimated [R.sup.2] can be computed as a measure of goodness of fit.(3)

VI. Empirical Results

Ordinal probit and ordinal logit models, with both measures of homogeneity, were estimated by the method of maximum likelihood. Numerical optimizations were carried out using GQOPT5,(4) with likelihood functions and gradients programmed in FORTRAN by the authors. The results are presented in Table I. The results for the ordinal logit and ordinal models were similar and the following conclusions can be drawn from either set of results. [Tabular Data I Omitted]

Brand-name recognition affects both the costs of running for office and the probability of winning the election, creating a brand-name barrier to entry in political markets that, as Lott [17; 18] has suggested, cannot be transferred. This leads to political production, the production of wealth transfers, by less efficient politicians.

Boundary changes alter the voter composition of a district. This may either increase or decrease competition in that district. The coefficient of the dummy variable BOUNDARY is negative, significant and stable across specifications, suggesting increased competition in those districts with changes in their borders because of the Boundary Act.

We find evidence that an election held in a borough is more likely to be contested than one held in a county district; the coefficient of BOROUGH is positive and significant in all models. This supports our hypothesis that the higher the cost of information, the larger the requirement for brand-name capital and the less competition for that district election. Candidates were willing to pay more to run in a district that, once won, would be a more secure seat in Parliament.

The inclusion of FAMILY tests only for the existence of the same last name in a previous election in the district, not necessarily family lineage (though often voters may use the same proxy, e.g., someone named Kennedy running for office in Massachusetts). Yet, the significant and negative coefficients for FAMILY in the various models supports Laband and Lentz's [16] and Lott's [17] contention that while brand-name capital may not be traded, it may be handed down from parent to child.

While the models generally perform well and the coefficients can be interpreted within the theoretical framework of political markets, some evidence is weak. For example, the coefficient of the variable, TITLED, has the correct sign, but is not significant in any of the models. The coefficient for our measure for tenure, VICTORIES, is negative but not significant. This variable measures not only incumbent investment in brand-name capital, but also incumbent efficiency.

We see that where we use a separate measure for homogeneity, either HOMOGENEITY or SIZE, greater homogeneity, by itself, results in less competition for the seat. If the local party organization has some cartel power, it will try to limit those running under its party banner to keep the costs down for the party. This is especially true with our party homogeneity measure.

Still, the two measures of homogeneity perform differently in the interaction with VICTORIES*BOROUGHS and DEFEATS*BOROUGHS. Using HOMOGENEITY with past campaigns, successful or not, the borough districts saw more competition in association with more past campaigns the more homogeneous they were than did the county districts that were similarly homogeneous. This is the result predicated by the theories of Nelson [24], Telser [29] and Crain and Goff [8].

Crain and Goff's measure, SIZE, leads to a contradictory result, as the coefficients are significantly positive, instead of negative as predicted. This suggests that Crain and Goff's measure of homogeneity may be problematic.

In this study, we test compare results from the ordinal probit and ordinal logit models. As the two models are obviously non-nested, we follow the likelihood-ratio test procedure of Vuong [31]. For convenience, denote the ordinal probit model as [F.sub.[Theta]], with log-likelihood (7) [Mathematical Expression Omitted] where f(.) is the contribution of observation t to the likelihood. Also, denote the ordinal logit model as [G.sub.[Tau]], with log-likelihood (8) [Mathematical Expression Omitted] We test the following hypotheses: [Mathematical Expression Omitted] where [E.sup.o] {.} denotes the expectation with respect to the true joint distribution of (y,x), and [[Theta].sub.*] and [[Tau].sub.*] are the pseudo-true values of [Theta] and [Tau], respectively [32]. Vuong [31]. has shown that, under the null hypothesis, (9) [Mathematical Expression Omitted] where [Mathematical Expression Omitted] and where [[Theta][Caret]] and [[Tau][Caret]] are the ML estimators for [Theta] and [Tau], respectively. To test the hypothesis, select a significance level [Alpha] and the critical value [z.sub.[Alpha]]. If z[Caret] > [z.sub.[Alpha], then [F.sub.[Theta]] is better than [G.sub.[Tau]] in that [F.sub.[Theta]] is closer to the true law generating the observations; if z[Caret] < -[z.sub.[Alpha]], then [F.sub.[Theta]] is worse than [G.sub.[Tau]]; if [Mathematical Expression Omitted], then [F.sub.[Theta] is not statistically different from [G.sub.[Tau]].

In the model where [H.sup.a] is SIZE, z[Caret] = 1.11, suggesting that the ordinal logit estimation performs better, but not significantly so, than the ordinal probit estimation. In the model where [H.sup.a] is HOMOGENEITY, z[Caret] = -2.51, suggesting that ordinal logit significantly outperforms ordinal probit at the 5% level.

VII. Conclusions

In the years before electronic media, information was passed from one person to the next in either the written or the spoken word. Information flowed at lower cost when people were physically closer to one another--in an urban instead of a rural environment. This lower cost of information in the urban district made it easier for a challenger to get elected, and so, more likely that incumbents would be challenged.

Informational capital limited competition in the parliamentary elections studied here. This supports Lott's contention [17; 18; 19] that inalienability of brand-name capital is an important factor in democratic processes. The large amount of campaign expenditures devoted to establishing a positive political image is largely motivated by an expectation of a high return on the investment. Once established, a high level of name recognition effectively deters potential competitors and secures a seat in the legislature.

Brand name can effectively bar entry only if it confers a significant advantage on the possessor and potential candidates recognize and appreciate the probable outcome of a contested election. If informational capital can be traded, there is no advantage to being an incumbent; more efficient challengers can purchase the political capital of incumbents, making elections unnecessary. Brand name gives incumbents a significant political advantage and potential competitors carefully evaluate the extent of this advantage before deciding to challenge. This rational decision mechanism reduced the degree of competition in Parliamentary elections.

Our results tend to support the view of Crain and Goff [8; 9] that the value of brand-name investment increases as information from other sources decreases. Our interpretation is that the apparent anomaly of much higher costs of standing for election in countries is mostly explained by the necessity for much greater investment in brand name. Significantly, while investment in brand name is valuable in both countries and boroughs, the value of the name as a signal t voters is much more important in the less densely populated countries where other information is generally more difficult to obtain. In counties, more than in boroughs, those with political brand names effectively exclude competitors, so that the value of the seat rises.

If successful candidates can transfer political capital to their children, candidates have an incentive to preserve that capital for their children, and there is less of a problem of candidates shirking in their last period in office. We find support for this hypothesis.

These results also support Nelson [26], Telser [29] and Crain and Goff [8; 9], that candidates are experience goods when information costs to voters are higher and search goods when these costs lower. We find that in the lower information-cost boroughs, additional campaigns, successful or not, held in less diverse districts (using our measure, HOMOGENEITY) builds political capital that is useful in future elections. Using Crain Goff's measure, SIZE, we get contradictory results, suggesting that either the theory or the measure is incorrect. (1) McCornick and Tollison argue that lack of pay is a barrier to entry [21, 96-97] Crain [5] discusses the effects of residency requirements on competition in political (2) To test the overall significance of the independent all slope coefficients equal to zero) reduces to [L [*] = [[Sigma] r.sub.k] log [r.sub.k] - n log n, where [r.sub.k] is the total number of observations in category k, and n is the sample size [23, 110]. Obviously, the likelihood-ratio test does not call for constrained ML estimation. (3) McKelvey and Zavoina [23, 111-12] suggests the following estimated [R[Caret].sup.2] as a goodness-of-fit measure: [Mathematical Expression Omitted] where [Mathematical Expression Omitted] is the ML estimator of [Beta], and n is the sample size. (4) The numerical optimization package GQOPT5 was prepared by Stephen Goldfeld and Richard Quandt. The FORTRAN codes for all estimation are available from the authors.

References

[1]Axelrod, Robert. The Evolution of Cooperation. New York: Basic Books, 1984. [2]Coats, Morris. "Voter Participation in Nineteenth Century British Parliamentary Elections." Ph.D. dissertation, Virginia Polytechnic Institute and State University, 1983, p. 14. [3]-- and Thomas Dalton, "Entry Barriers in Politics and Uncontested Elections." Journal of Public Economics, forthcoming. [4]Craig, F. W. S. British Parliamentary Election Results: 1832-1885. London: MacMillan, 1977. [5]Crain, Mark, "On the Structure and Stability of Political Markets." Journal of Political Economy, August 1977, 829-42. [6]-- and Robert D. Tollison, "Campaign Expenditures and Political Competition." Journal of Law and Economics, April 1976, 177-88. [7]--, Thomas Deaton, and Robert D. Tollison, "Legislators as Taxi Cabs: On the Value of Seats in the U.S. House of Representatives." Economic Inquiry, April 1977, 298-302. [8]-- and Brian Goff, "Televising Legislatures: An Economic Analysis." Journal of Law and Economics, October 1986, 405-21. [9]-- and --. Televised Legislatures: Political Information, Technology and Public Choice. Boston: Kluwer, 1988. [10]Demsetz, Harold, "Barriers to Entry." American Economic Review, March 1982, 47-57. [11]Ferguson, James M., "Comment on Political Information." Journal of Law and Economics, August 1976, 341-46. [12]Fremling, Gertrude and John Lott, "Televising Legislatures: Some Thoughts on Whether Politicians Are Search Goods." Public Choice, vol. 58, no. 1, 1988, 73-78. [13]Goldfeld, Stephen M. and Richard E. Quandt. Nonlinear Methods in Econometrics. Amsterdam: North-Holland, 1972. [14]Gourieroux, C., A. Monfort, and A. Trognon, "Pseudo-maximum Likelihood Methods: Applications to Poisson Models." Econometrica, May 1984, 701-20. [15]Klein, Benjamin and Keith B. Leffler, "The Role of Market Forces in Assuring Contractual Performance." Journal of Political Economy, August 1981, 615-41. [16]Laband, David N. and Bernard F. Lentz, "Favorite Sons: Intergenerational Wealth Transfers Among Politicians." Economic Inquiry, July 1985, 395-414. [17]Lott, John R., "Brand Names and Barriers to Entry in Political Markets." Public Choice, 1986, 87-92. [18]--, "The Effects of Nontransferrable Property Rights on the Efficiency of Political Markets: Some Evidence." Journal of Public Economics, March 1987, 231-46. [19]--, "Licensing and Nontransferrable Rents." American Economic Review, June

1987, 453-55. [20]Maddala, G. S. Limited-dependent and Qualitative Variables in Econometrics. Cambridge: University Press, 1983. [21]McCormick, Robert and Robert D. Tollison. Politicians, Legislation, and the Economy. Boston: Martinus-Nijhoff Publishers,1981. [22]McCullagh, P. and J. A. Nelder. Generalized Linear Models. New York: Chapman & Hall, 1983. [23]McKelvey, Richard D. and William Zavoina, "A Statistical Model for the Analysis of Ordinal Level Dependent Variables." Journal of Mathematical Sociology, December 1975, 103-20. [24]Nelson, Phillip, "Information and Consumer Behavior." Journal of Political Economy, March/April 1970, 311-29. [25]--, "Advertising As Information." Journal of Political Economy, July/August 1974, 729-54. [26]--, "Political Information." Journal of Law and Economics, August 1976, 315-36. [27]Pinto-Duchinsky, Michael. British Political Finance: 1830-1980. Washington: American Enterprise Institute, 1981. [28]Pratt, John W., "Concavity of the Log-likelihood." Journal of the American Statistical Association, March 1981, 103-106. [29]Telser, Lester G., "Comment of Political Information." Journal of Law and Economics, August 1976, 337-40. [30]Tullock, Gordon, "Entry Barriers in Politics." American Economic Review May, 1965, 458-66. [31]Vuong, Quang H., Likelihood Ratio Tests for Model Selection and Non-nested Hypotheses." Econometrica, March 1989, 307-33. [32]White, Halbert, "Maximum Likelihood Estimation of Misspecified Models." Econometrica, January 1982, 1-25.

Advertising is often viewed as a sunk investment that will receive a return only if firms do not cheat their customers on quality, much like a security bond that will be lost for non-performance of a contract. In the last period a firm will cheat if there is no future return from the investment [15]. Knowing that a firm is in it last period will then reduce sales in the last period to zero, so the firms will begin to cheat the period before the last. This is the usual iterated prisoner's dilemma (PD) game with a finite life; cheating begins in the first period and continues. Demsetz [10] argues that since firms are transferable through voluntary trade, the possibly infinite life of a firm transforms this finite, iterated PD game with a known end into an iterated PD game with no known end. As Axelrod [1] suggests, the best of the known strategies for this game is the tit-for-tat strategy. In this game, firms live up to quality claims if customers continue to pay for their product. Demsetz also argues that brand name cannot serve as a barrier to entry for alienable firms because the firm, with its brand name intact, can be purchased by buyers who are more efficient producers than the current producer.

Since professionals and politicians who have made investments in brand name have namebrand capital that is inalienable, these markets may suffer from the last-period problem [12; 17; 18; 19]. If the seller knows the last period because he intends to quit, cheating cannot be prevented by investment in brand-name capital. Still, uncertain ending periods may be enough to insure that the seller will produce at the advertised quality if buyers determine the final period [29]. V

Lott [18] has incorporated brand-name entry barriers into a model analyzing competition for legislative seats. Though Lott fundamentally accepts Demsetz's [10] critique of entry barrier analysis, Lott points out that Demsetz's reasoning does not hold in cases where the producer is not an ongoing enterprise, where the producer is a single individual, not a firm. Political brand-name investment erects a barrier to entry because this capital is not transferable.

Nelson [24; 25] distinguishes between the "search" and "experience" qualities in a good. Search qualities are those that can be discovered in the search process before purchase, and experience qualities are those that can be discovered only after purchase, as the product is used or consumed.

Nelson [26] argues that the repeat-purchase mechanism gives candidates some incentive to be truthful in their advertising. Further, the candidate has an incentive to direct his advertisement to those who would most likely vote for him, if they knew his position. Telser [29] and Ferguson [11] suggest that candidates are experience goods and so the voter knows that he cannot trust the politician to be completely honest and forthcoming. The voter instead relies on experience with that candidate (brand) or relies on trustworthy judges of candidates that act as agents for voters, i.e., the political party allowing its brand name to be used by the candidate.

The incumbent's advantage stems partially from his fund-raising advantage, which can be translated into advertising that builds brand-name capital. This fund-raising advantage cannot be translated into an advertising advantage for a candidate in a politically diverse district if the candidate is a search good [8; 12; 29]. These candidates' positions are well known and a greater amount of funds raised is viewed as a greater debt owed by the candidate to some supporting group: a group which some will oppose in diverse districts. Advertising becomes a signal about the level of debt owed to some interest group, which signals a worse buy; not a better buy, as is true if the candidate were an experience good. In homogeneous districts, investment in brand V name, by making the candidate's positions known to the constituency increases his expected vote margin and decreases the chance that a rival will come forward [8; 12; 29]. On the other hand, such investment in heterogeneous districts increases both the support and the opposition to the candidate.

Since the difference between search and experience goods depends on the cost of measuring the quality of a good before and after purchase, this cost difference can be changed by new technology and new institutions. Crain and Goff [8] examine the televising of legislatures, which is found in many, but not all, states. Crain and Goff postulate that politicians will be search goods in the states where legislatures are widely televised but will experience goods where they are seldom televised. They test their hypothesis using 1976 state house and senate races by stratifying their sample into high television exposure states and low television exposure states. Crain and Goff use population per seat as a measure of political diversity, reasoning that larger populations will be more diverse. To test their hypothesis, they regress population per seat, the number of multi-member districts, and seats up for election against number of winning challengers. They find that as population per seat increases, challengers win more often in high exposure states than in low exposure states. They interpret this result as support for the hypothesis that politicians are experience goods where the cost of information is high and search goods where the information costs are low.

Fremling and Lott [12] argue that politicians are not experience goods since investment in brand name will not prevent cheating due to the existence of the last-period problem. This prevents conspicuous investment in brand name from serving as a guarantee that the politician will not act contrary to his public political promises. They argue instead that politicians are search goods because voters learn the politician's utility function before any expected last period.

Brand name is not the only factor that may discourage challengers from coming forward in an election. More efficient production by a provider may keep potential entrants our of a market. This type of advantage and barrier is welfare enhancing. One example of this type of barrier is the advantage of tenure of the candidate that is not due to his sunk investment but to his investment in human capital through experience on the job. Other advantages, which we associate with brand name, may be due less to the candidate's efficiency and more to his sunk investment in goodwill.

In this paper we empirically explore the brand-name barriers to entry in British Parliamentary contests. Political entry barriers have previously been examined [6; 7; 16; 30]. We examine the effect on competition of important measures of brand-name investment identified elsewhere [3] and include measures of political diversity. Our measure of competitiveness is the excess of the number of candidates over the number of seats available. In the next section, we examine the consequences of information-cost differentials between British rural districts and urban districts in the last century. The third section discusses political diversity and brand-name advantage. Then, the data set is described, and the variables used in the empirical section are discussed. The fifth section presents a discussion of ordinal probit and ordinal logit models used to analyze the data and test our hypothesis. The results of the ordinal response models are discussed in the sixth section, including a test between the ordinal probit and ordinal logit specifications for our model.

II. The Cost of Standing and the Cost of Information

In Britain from 1832, the average cost of standing for a county seat was roughly 20,000 pounds in 1980 pounds, while a borough seat cost only about one-half that much [27]. Candidates paid more to run for county seats than for borough seats, though there were no residency requirements and county Members of Parliament (M.P.'s) lacked greater legislative powers than the borough M.P.'s. Also, neither the county M.P. nor the borough M.P. was paid a salary.(1) Why did candidates consistently pay twice the price of a borough seat to present a county district? One explanation is that the county seat was more secure than the borough seat [2]. The rural seat was more valuable because it was easier to retain once won.

Greater political security in the county seats can be explained by the higher cost of information in the rural districts. In the more densely populated urban districts, there was more V interaction with a greater variety of people, information spread by word of mouth more quickly and at lower cost. Following Crain and Goff [8], politicians are expected to be experience goods in countries and search goods in boroughs. Being an incumbent acts as a signal, a low-cost information source about the candidate. Investment in political brand name as campaign expenditure acts as a barrier to entry for the high information-cost county seats (though the real barrier to entry is the high-cost information). In the borough districts, lower information costs reduce incumbent advantage and the advantage from previous investment in brand name.

Telser [29], Nelson [26], and Crain and Goff [8] suggest that the degree of political diversity of the electorate influences the advantage of political brand-name investment. When politicians are search goods, as we expect them to be in boroughs, investment in brand name will be of little value in more diverse districts since politicians' positions are easily discovered without such investment and such investment suggests a debt owed to some special interest group. When politicians are experience goods, as we except them to be in counties, increasing political diversity will not decrease the value of brand-name investment since brand name informs the voters about otherwise unknown positions.

Different measures for political diversity can be used. Crain and Goff [8] use population per seat. A weakness of this measure is its questionable assumption that districts of the same size have the same degree of diversity, or that diversity increase with size. We propose an alternate measure, party homogeneity, which measures the diversity in part vote shares in recent elections in a districts. We use both Crain and Goff's measure and our own in testing the importance of brand-name investment in more versus less diverse districts and higher versus lower information-cost districts.

The hypothesis that we text requires some explanation. Since investment in brand-name capital in more homogeneous districts gives the candidate greater advantage than the same investment in a more diverse district, the interaction of brand-name investment variables with some measure of diversity should be negative (or positive with a measure of homogeneity). Furthermore, the slope of entry difficulty with respect to this interaction of diversity and past investment in brand name should be more negative in the low information-cost boroughs than the high information-cost countries. Information costs are hypothesized to be greater in the county districts than in the borough districts, leading us to expect greater importance in brand-name investment in reducing competition in the county districts. Investment in non-transferable brand-name capital, as

V measured both by past campaign victories and past campaign losses, should be more important in reducing the number of competitors in the county districts than in the borough districts. Our null hypothesis, then, is that the coefficient of the interaction of district diversity and past investment in brand-name will be no less in the lower information-cost boroughs than that coefficient for the county districts. Once we control for diversity with either measure (included separately in the regression), we can examine these interactions.

IV. The Data Set

Just as institutional and technological differences can lead to differences in the cost of pre-purchase information concerning candidate quality, sociological differences also can lead to such information-cost differentials. The difference explored in this paper is the difference between

V urban districts and rural districts before the coming of television and radio.

Crain and Goff [8] stratify their sample of states by high and low television exposure of the legislature, which effectively stratifies by cost of pre-purchase information about candidate quality. We stratify our sample by cost of information in another way, that is, by urban versus county districts. We postulate that in nineteenth century Britain, political information was transmitted either by talk or by print. Since it is less costly to talk with others and to read newspapers and pamphlets, we conclude that information costs are lower in the urban or borough districts than in the rural or county districts.

Since our original question is about the differential cost of standing in urban and rural districts in nineteenth century Britain, we have chosen British Parliamentary general elections from 1852 to 1880 as the population from which to draw our sample. Craig [4] has compiled results from all 381 districts in England, Wales, Scotland, and the Universities, leaving out the Irish constituencies from 1832 to 1885. We begin our data set with 1852 so that we can compile at least a twenty-year history of the candidates in that district before the first election in our data set. Further, we have taken a stratified sample of roughly sixty percent of English boroughs and counties, Scottish boroughs and countries, and Welsh boroughs and countries that were enfranchised in 1832. We have left out the five University districts because only two of these were enfranchised in 1832.

There are several features of this data set that make it especially useful for our purposes. First, it includes one easy-to-measure brand-name factor that can influence British Parliamentary elections, title of nobility. We also obtain a proxy for family electoral experience in that district because the data set includes full names and vote totals for all candidates during this period. Finally, the data set contains multi-seat districts with anywhere from one to three seats.

Economic theory predicts that it is more desirable for a candidate to contest an election (and therefore, there will be more candidates) when the rewards of office are higher, when the probability of winning is higher, or when the costs of challenging are lower. Our data set does not allow us to examine directly the issue of returns to office holding. Still, we can examine the strategies of individuals faced with an uncertain election result. Potential candidates generally are expected to have poor information regarding other potential non-incumbent candidates' chances of winning, and largely discount the effect of these candidates. We except that potential candidates are likely to focus on the current incumbents who are seeking re-election. The problem facing potential candidates is to decide whether to complete by judging their chances against the chances of incumbent-candidates.

A candidate's initial brand-name capital, the stock of information about the candidate in the hands of the voters in that district before the election gets under way, increases that candidate's chance of winning. If that candidate is an incumbent, the initial stock of brand-name capital also reduces the chances of other candidates and so should reduce the number of candidates opposing the incumbent. If the incumbent has won several elections in the past, has the family name of a former M.P. from the district, has run in several past elections albeit unsuccessfully, or has a title of nobility, the information about the incumbent candidate is greater at the beginning of the campaign, so initial brand-name capital of that incumbent candidate is greater. The main variables in the empirical model measure initial brand-name capital.

The observational unit for this study is the district election, not the seat. The data set contains variables by district and year for 200 districts

V and the election years from 1852-80. The variables are:

1. LOSERS: the dependent variable. This variable is the number of candidates in excess of the number of seats available, or the number of losing candidates. The values for LOSERS range from zero to four. An election is considered to be more competitively fought by the extent the number of candidates exceeds the number of seats.

2. VICTORIES: the total number of elections incumbent-candidates have won in this district. Previous successful campaign should increase name recognition and one's electoral strength in subsequent elections. Still, tenure also increases experience, and so, the efficiency in income redistribution activities.

3. DEFEATS: the total number of times incumbent-candidates have run unsuccessfully in this district. Such office seeking may contribute to name recognition, increasing one's chances in subsequent elections. Whereas VICTORIES measures both investment in brand-name capital and efficiency, this variable measures only investment in brand-name capital.

4. TITLED: the number of incumbent-candidates who are titled. Since title has brand name connotations, this variable should be positively related to LOSERS. Lott [17] states that brand name acquired outside of political markets may be valuable in a political market.

5. FAMILY: the number of incumbent-candidates who have the same last name as previous winning candidates in the district. Laband and Lentz [16] and Lott [17] suggest that while brand name cannot be traded, it can be handed down. This is way family name is considered important in this study.

6. BOROUGH: a dummy variable with a value of one if the election is in a borough and zero V if it is in a county.

7. BOUNDARY: a dummy variable to capture the effect of the 1868 Parliamentary Boundary Act. This Act reapportioned the legislative districts, adding new districts, combining districts and eliminating some districts [4]. New districts added by the Act were excluded from our sample because we need the history of participation from 1832-52 to create several independent variables. We include the dummy to capture the effect of this redistricting on competition. We could not control for other boundary changes that may have occurred occasionally throughout the period

V of study.

8. WALES and ENGLAND: two dummy variables to account for possible differences in competitiveness in the three countries.

9. FIFTY-SEVEN, FIFTY-NINE, SIXTY-FIVE, SIXTY-EIGHT, SEVENTY-FOUR and EIGHTY: dummy variables for the election years. 1857, 1859, 1865, 1868, 1874, and 1880, respectively. These variables are used to capture differences among general elections (year of the general election).

The next two independent variables, party homogeneity (HOMOGENEITY) and electorate size per seat (SIZE), measure the homogeneity or the lack it, of the district. Crain and Goff's measure assumes that larger electorates are less homogeneous than smaller districts [8]. We believe, however, the Crain and Goff's measure of homogeneity is unsatisfactory since in particular data sets, such as ours, certain large districts may be more homogeneous than smaller districts. Our alternate measure, HOMOGENEITY, is calculated as: (1) [Mathematical Expression Omitted] where LIBERAL [VOTE.sub.t] is the total votes for liberal candidates in election r, and TOTAL [VOTE.sub.t] is the total votes in election t. The variable HOMOGENEITY will range from zero to fifty, such that HOMOGENEITY plus fifty is the average percent of the votes cast over the past four elections for the predominant party.

Using this data set in previous study, Coats and Dalton [3] found that the variables VICTORIES, DEFEATS, TITLED, and FAMILY were correlated with whether and election was contested. If a district were not contested, the barriers to entry were too high for challengers to overcome, but if an election were held, any entry barriers were insufficient to forestall competitors. We believe that LOSERS measures the degree of competition in an election better than a dichotomous variable that is based on contested versus uncontested elections. As the number of candidates increased, holding the number of seats constant, entry barriers in the district fell as hurdle for challengers. Therefore, LOSERS is a measure of the size of the barriers to entry, or the degree of competition for a seat.

V. The Empirical Models

We estimate the following model, using two different measures of homogeneity, HOMOGENEITY and SIZE, and using two different econometric specifications, ordinal probit and ordinal logit: (2) LOSERS = f (FAMILY, TITLED, VICTORIES, DEFEATS, ENGLAND,

WALES, BOUNDARY, BOROUGH, [H.sup.a], VICTORIES * [H.sup.a], DEFEATS * [H.sup.a], VICTORIES * [H.sup.a] * BOROUGH, DEFEATS * [H.sup.a] * BOROUGH, FIFTY -- SEVEN, FIFTY -- NINE, SIXTY -- FIVE, SIXTY -- EIGHT, SEVENTY-FOUR and EIGHTY),

where [H.sup.a] is HOMOGENEITY in one case and SIZE in the other.

The homogeneity (or heterogeneity) variables are included in the models alone and in interaction terms with VICTORIES and DEFEATS (measures of past investment in brand name capital and signals of electoral strength). The theory suggests that the more homogeneous the district, the greater the impact of brand-name investment on reducing competition, since brand-name signals a better buy for the voters in these districts. In higher information-cost districts, such as our county districts and Crain and Goff's [8] low televised legislature states, this brand name signal should be more important. We include homogeneity by itself since we except that districts with a more even split between party allegiances of voters will be more fertile ground for opposition parties to make inroads and should be more highly contested districts.

The variables, DEFEATS and VICTORIES, provide measures of previous investment in political brand name and are used to examine Lott's [17] brand-name hypothesis. Though VICTORIES measures both efficiency in arranging transfer and previous investment in brand-name capital, DEFEATS measures only previous investment in brand-name capital. The variable BOROUGH and the BOROUGH dummy-interaction variables, BOROUGH*VICTORIES, and BOROUGH*DEFEATS are used to examine the hypothesis of differential election cost developed in this paper. The other variables are included to control for other effects over the various districts and election years.

The major consideration in the econometric estimation is the special feature of the data. In particular, the sample space of the dependent variable is the set of nonnegative integers. Here, standard econometric techniques that do not consider this special feature of the data, such as ordinary least-squares regression, present well-known problems [14; 20].

The recent developments in the analysis of count data have contributed significantly in this aspect of econometric modeling. One approach to the analysis of count data is through the estimation of nonlinear regression [14; 22]. But, these models do not consider the discrete and nonnegative nature of the count data. Another approach is through the specification of certain probability distributions such as Poisson distribution. The other approach, adopted in this study, is to treat the dependent variable as ordinal. To this end, McKelvey and Zavoina [23] have developed one form categorical model, known as the response model.

In the following we present the ordinal response model as it appears in the literature. Denote the response function as (3) [Mathematical Expression Omitted] where for observation i, [Mathematical Expression Omitted] is the theoretical response variable of interest, [x.sub.i] is a vector of explanatory variables, [Beta] is a vector of unknown parameters, and [u.sub.i] is a random disturbance term with E([u.sub.i/x.sub.i]) = 0 for all i. Then, [Mathematical Expression Omitted] for all i. Although [Mathematical Expression Omitted] is not observed, it belongs to a known category associated with an integer value for the observed variable [y.sub.i]. Following McKelvey and Zavoina [23], define a vector set of m + 1 constants [Mu] = ([[Mu].sub.o], [[Mu].sub.1,...,[[Mu.]sub.m]), such that [Mathematical Expression Omitted], where m is the number of categories. For the ith observation, [Mathematical Expression Omitted] belongs to the jth category if and only if (4) [Mathematical Expression Omitted] Also, define a dichotomous variable [Z.sub.ij] = 1 if [Mathematical Expression Omitted] falls in the jth category; = 0 otherwise. Then, for an observation i, condition (4) translates into the following probability in reference to (3): (5) [Mathematical Expression Omitted] where F is a cumulative distribution function. For a sample of size n, the likelihood function is (6) [Mathematical Expression Omitted]

If the distribution function F is specified as the cumulative standard normal function, equation (6) is the likelihood function for the ordinal probit model developed in McKelvey and Zavoina [23] and presented in Maddala [20, 46-49]. If F is specified as the logistic function, then (6) is the likelihood function for the ordinal logit model. In either case, the maximum likelihood (ML) method can be applied in the estimation of the slope coefficients [Beta] and the ancillary parameters [Mu]. Pratt [28] has shown that the log-likelihood functions for the ordinal probit model and the ordinal logit model are globally concave. Therefore, for both models the ML estimates are unique. For statistical inferences, like most other ML estimation, inverting the information matrix yields the variance-covariance matrix of the ML estimates that can be used for tests of significance [13]. To find the significance of all or a subset of coefficients, likelihood-ratio tests can be performed based on the log-likelihood function values.(2) In addition, an estimated [R.sup.2] can be computed as a measure of goodness of fit.(3)

VI. Empirical Results

Ordinal probit and ordinal logit models, with both measures of homogeneity, were estimated by the method of maximum likelihood. Numerical optimizations were carried out using GQOPT5,(4) with likelihood functions and gradients programmed in FORTRAN by the authors. The results are presented in Table I. The results for the ordinal logit and ordinal models were similar and the following conclusions can be drawn from either set of results. [Tabular Data I Omitted]

Brand-name recognition affects both the costs of running for office and the probability of winning the election, creating a brand-name barrier to entry in political markets that, as Lott [17; 18] has suggested, cannot be transferred. This leads to political production, the production of wealth transfers, by less efficient politicians.

Boundary changes alter the voter composition of a district. This may either increase or decrease competition in that district. The coefficient of the dummy variable BOUNDARY is negative, significant and stable across specifications, suggesting increased competition in those districts with changes in their borders because of the Boundary Act.

We find evidence that an election held in a borough is more likely to be contested than one held in a county district; the coefficient of BOROUGH is positive and significant in all models. This supports our hypothesis that the higher the cost of information, the larger the requirement for brand-name capital and the less competition for that district election. Candidates were willing to pay more to run in a district that, once won, would be a more secure seat in Parliament.

The inclusion of FAMILY tests only for the existence of the same last name in a previous election in the district, not necessarily family lineage (though often voters may use the same proxy, e.g., someone named Kennedy running for office in Massachusetts). Yet, the significant and negative coefficients for FAMILY in the various models supports Laband and Lentz's [16] and Lott's [17] contention that while brand-name capital may not be traded, it may be handed down from parent to child.

While the models generally perform well and the coefficients can be interpreted within the theoretical framework of political markets, some evidence is weak. For example, the coefficient of the variable, TITLED, has the correct sign, but is not significant in any of the models. The coefficient for our measure for tenure, VICTORIES, is negative but not significant. This variable measures not only incumbent investment in brand-name capital, but also incumbent efficiency.

We see that where we use a separate measure for homogeneity, either HOMOGENEITY or SIZE, greater homogeneity, by itself, results in less competition for the seat. If the local party organization has some cartel power, it will try to limit those running under its party banner to keep the costs down for the party. This is especially true with our party homogeneity measure.

Still, the two measures of homogeneity perform differently in the interaction with VICTORIES*BOROUGHS and DEFEATS*BOROUGHS. Using HOMOGENEITY with past campaigns, successful or not, the borough districts saw more competition in association with more past campaigns the more homogeneous they were than did the county districts that were similarly homogeneous. This is the result predicated by the theories of Nelson [24], Telser [29] and Crain and Goff [8].

Crain and Goff's measure, SIZE, leads to a contradictory result, as the coefficients are significantly positive, instead of negative as predicted. This suggests that Crain and Goff's measure of homogeneity may be problematic.

In this study, we test compare results from the ordinal probit and ordinal logit models. As the two models are obviously non-nested, we follow the likelihood-ratio test procedure of Vuong [31]. For convenience, denote the ordinal probit model as [F.sub.[Theta]], with log-likelihood (7) [Mathematical Expression Omitted] where f(.) is the contribution of observation t to the likelihood. Also, denote the ordinal logit model as [G.sub.[Tau]], with log-likelihood (8) [Mathematical Expression Omitted] We test the following hypotheses: [Mathematical Expression Omitted] where [E.sup.o] {.} denotes the expectation with respect to the true joint distribution of (y,x), and [[Theta].sub.*] and [[Tau].sub.*] are the pseudo-true values of [Theta] and [Tau], respectively [32]. Vuong [31]. has shown that, under the null hypothesis, (9) [Mathematical Expression Omitted] where [Mathematical Expression Omitted] and where [[Theta][Caret]] and [[Tau][Caret]] are the ML estimators for [Theta] and [Tau], respectively. To test the hypothesis, select a significance level [Alpha] and the critical value [z.sub.[Alpha]]. If z[Caret] > [z.sub.[Alpha], then [F.sub.[Theta]] is better than [G.sub.[Tau]] in that [F.sub.[Theta]] is closer to the true law generating the observations; if z[Caret] < -[z.sub.[Alpha]], then [F.sub.[Theta]] is worse than [G.sub.[Tau]]; if [Mathematical Expression Omitted], then [F.sub.[Theta] is not statistically different from [G.sub.[Tau]].

In the model where [H.sup.a] is SIZE, z[Caret] = 1.11, suggesting that the ordinal logit estimation performs better, but not significantly so, than the ordinal probit estimation. In the model where [H.sup.a] is HOMOGENEITY, z[Caret] = -2.51, suggesting that ordinal logit significantly outperforms ordinal probit at the 5% level.

VII. Conclusions

In the years before electronic media, information was passed from one person to the next in either the written or the spoken word. Information flowed at lower cost when people were physically closer to one another--in an urban instead of a rural environment. This lower cost of information in the urban district made it easier for a challenger to get elected, and so, more likely that incumbents would be challenged.

Informational capital limited competition in the parliamentary elections studied here. This supports Lott's contention [17; 18; 19] that inalienability of brand-name capital is an important factor in democratic processes. The large amount of campaign expenditures devoted to establishing a positive political image is largely motivated by an expectation of a high return on the investment. Once established, a high level of name recognition effectively deters potential competitors and secures a seat in the legislature.

Brand name can effectively bar entry only if it confers a significant advantage on the possessor and potential candidates recognize and appreciate the probable outcome of a contested election. If informational capital can be traded, there is no advantage to being an incumbent; more efficient challengers can purchase the political capital of incumbents, making elections unnecessary. Brand name gives incumbents a significant political advantage and potential competitors carefully evaluate the extent of this advantage before deciding to challenge. This rational decision mechanism reduced the degree of competition in Parliamentary elections.

Our results tend to support the view of Crain and Goff [8; 9] that the value of brand-name investment increases as information from other sources decreases. Our interpretation is that the apparent anomaly of much higher costs of standing for election in countries is mostly explained by the necessity for much greater investment in brand name. Significantly, while investment in brand name is valuable in both countries and boroughs, the value of the name as a signal t voters is much more important in the less densely populated countries where other information is generally more difficult to obtain. In counties, more than in boroughs, those with political brand names effectively exclude competitors, so that the value of the seat rises.

If successful candidates can transfer political capital to their children, candidates have an incentive to preserve that capital for their children, and there is less of a problem of candidates shirking in their last period in office. We find support for this hypothesis.

These results also support Nelson [26], Telser [29] and Crain and Goff [8; 9], that candidates are experience goods when information costs to voters are higher and search goods when these costs lower. We find that in the lower information-cost boroughs, additional campaigns, successful or not, held in less diverse districts (using our measure, HOMOGENEITY) builds political capital that is useful in future elections. Using Crain Goff's measure, SIZE, we get contradictory results, suggesting that either the theory or the measure is incorrect. (1) McCornick and Tollison argue that lack of pay is a barrier to entry [21, 96-97] Crain [5] discusses the effects of residency requirements on competition in political (2) To test the overall significance of the independent all slope coefficients equal to zero) reduces to [L [*] = [[Sigma] r.sub.k] log [r.sub.k] - n log n, where [r.sub.k] is the total number of observations in category k, and n is the sample size [23, 110]. Obviously, the likelihood-ratio test does not call for constrained ML estimation. (3) McKelvey and Zavoina [23, 111-12] suggests the following estimated [R[Caret].sup.2] as a goodness-of-fit measure: [Mathematical Expression Omitted] where [Mathematical Expression Omitted] is the ML estimator of [Beta], and n is the sample size. (4) The numerical optimization package GQOPT5 was prepared by Stephen Goldfeld and Richard Quandt. The FORTRAN codes for all estimation are available from the authors.

References

[1]Axelrod, Robert. The Evolution of Cooperation. New York: Basic Books, 1984. [2]Coats, Morris. "Voter Participation in Nineteenth Century British Parliamentary Elections." Ph.D. dissertation, Virginia Polytechnic Institute and State University, 1983, p. 14. [3]-- and Thomas Dalton, "Entry Barriers in Politics and Uncontested Elections." Journal of Public Economics, forthcoming. [4]Craig, F. W. S. British Parliamentary Election Results: 1832-1885. London: MacMillan, 1977. [5]Crain, Mark, "On the Structure and Stability of Political Markets." Journal of Political Economy, August 1977, 829-42. [6]-- and Robert D. Tollison, "Campaign Expenditures and Political Competition." Journal of Law and Economics, April 1976, 177-88. [7]--, Thomas Deaton, and Robert D. Tollison, "Legislators as Taxi Cabs: On the Value of Seats in the U.S. House of Representatives." Economic Inquiry, April 1977, 298-302. [8]-- and Brian Goff, "Televising Legislatures: An Economic Analysis." Journal of Law and Economics, October 1986, 405-21. [9]-- and --. Televised Legislatures: Political Information, Technology and Public Choice. Boston: Kluwer, 1988. [10]Demsetz, Harold, "Barriers to Entry." American Economic Review, March 1982, 47-57. [11]Ferguson, James M., "Comment on Political Information." Journal of Law and Economics, August 1976, 341-46. [12]Fremling, Gertrude and John Lott, "Televising Legislatures: Some Thoughts on Whether Politicians Are Search Goods." Public Choice, vol. 58, no. 1, 1988, 73-78. [13]Goldfeld, Stephen M. and Richard E. Quandt. Nonlinear Methods in Econometrics. Amsterdam: North-Holland, 1972. [14]Gourieroux, C., A. Monfort, and A. Trognon, "Pseudo-maximum Likelihood Methods: Applications to Poisson Models." Econometrica, May 1984, 701-20. [15]Klein, Benjamin and Keith B. Leffler, "The Role of Market Forces in Assuring Contractual Performance." Journal of Political Economy, August 1981, 615-41. [16]Laband, David N. and Bernard F. Lentz, "Favorite Sons: Intergenerational Wealth Transfers Among Politicians." Economic Inquiry, July 1985, 395-414. [17]Lott, John R., "Brand Names and Barriers to Entry in Political Markets." Public Choice, 1986, 87-92. [18]--, "The Effects of Nontransferrable Property Rights on the Efficiency of Political Markets: Some Evidence." Journal of Public Economics, March 1987, 231-46. [19]--, "Licensing and Nontransferrable Rents." American Economic Review, June

1987, 453-55. [20]Maddala, G. S. Limited-dependent and Qualitative Variables in Econometrics. Cambridge: University Press, 1983. [21]McCormick, Robert and Robert D. Tollison. Politicians, Legislation, and the Economy. Boston: Martinus-Nijhoff Publishers,1981. [22]McCullagh, P. and J. A. Nelder. Generalized Linear Models. New York: Chapman & Hall, 1983. [23]McKelvey, Richard D. and William Zavoina, "A Statistical Model for the Analysis of Ordinal Level Dependent Variables." Journal of Mathematical Sociology, December 1975, 103-20. [24]Nelson, Phillip, "Information and Consumer Behavior." Journal of Political Economy, March/April 1970, 311-29. [25]--, "Advertising As Information." Journal of Political Economy, July/August 1974, 729-54. [26]--, "Political Information." Journal of Law and Economics, August 1976, 315-36. [27]Pinto-Duchinsky, Michael. British Political Finance: 1830-1980. Washington: American Enterprise Institute, 1981. [28]Pratt, John W., "Concavity of the Log-likelihood." Journal of the American Statistical Association, March 1981, 103-106. [29]Telser, Lester G., "Comment of Political Information." Journal of Law and Economics, August 1976, 337-40. [30]Tullock, Gordon, "Entry Barriers in Politics." American Economic Review May, 1965, 458-66. [31]Vuong, Quang H., Likelihood Ratio Tests for Model Selection and Non-nested Hypotheses." Econometrica, March 1989, 307-33. [32]White, Halbert, "Maximum Likelihood Estimation of Misspecified Models." Econometrica, January 1982, 1-25.

Printer friendly Cite/link Email Feedback | |

Author: | Dalton, Thomas R. |
---|---|

Publication: | Southern Economic Journal |

Date: | Apr 1, 1992 |

Words: | 6117 |

Previous Article: | The substitutability of monetary assets in Taiwan. |

Next Article: | Entry, collusion, and capacity constraints. |

Topics: |