the locational determinants of western nonmetro high tech manufacturers

The Locational Determinants of WesternNonmetro High Tech Manufacturers:An Econometric Analysis

David L. Barkley and John E. Keith

The Tobit estimation procedure was used to determine the factors which influence thelocation and size of high technology manufacturers in nonmetro areas in the West.The results indicate that high tech branch plants tend to locate in populous countiesadjacent to Metropolitan Statistical Areas (MSAs). Percent of local employment inmanufacturing and agriculture was inversely related to branch plant employment, andthe stock of human capital was not significantly related to employment. High techunit plants also exhibited a propensity to locate in the more populous counties.Unlike branch plants, the unit concerns were more likely to develop or locate incommunities with a highly educated work force and at greater distances from metroareas. The unit plants better fit the perception of high tech plants selecting highamenity locations with abundant skilled labor.

Key words: high tech manufacturing, location, nonmetropolitan, qualitative models.

Declining employment in resource-based ac-tivities and mature manufacturing industrieshas encouraged nonmetropolitan communi-ties to seek alternative sources of basic em-ployment and income. Following the lead ofmetropolitan areas, many rural communitieshave investigated the possibility of attractingor generating employment in the rapidly grow-ing and skilled labor-intensive high technologymanufacturing industries with some recent ev-idence of success. High technology industriesare decentralizing their manufacturing activi-ties, and as a result, nonmetropolitan employ-ment growth in this sector has increased (Mar-kusen, Hall, and Glasmeier; Barkley;Glasmeier; Miller 1989). For example, Barkleyestimated that in 1982 over 511,000 high tech

David L. Barkley is a professor and an economic developmentspecialist in the Department of Agricultural and Applied Econom-ics, Clemson University; John E. Keith is a professor in the Eco-nomics Department, Utah State University. Seniority of author-ship is not implied by the order of authors' names.

This research was supported in part by the Western Rural De-velopment Center, Oregon State University, and the Utah andArizona Agricultural Experiment Stations, Regional Project W-165.

The authors gratefully acknowledge helpful comments from theeditor, Bruce Weber, and anonymous reviewers. Any remainingerrors or omissions are, of course, the responsibility of the authors.

manufacturing jobs were located in nonmetro-politan counties. Moreover, during the period1975-82, the nonmetropolitan employmentgrowth rate in this sector exceeded 15%. 1

Nonmetropolitan areas in the West havebeen successful in attracting or generating em-ployment in high technology manufacturing.Over 27,000 high tech manufacturing jobs werelocated in the nonmetro West in 1982, and the1975-82 growth rate exceeded 90% (Barkley,Keith, and Smith). This employment in hightech manufacturing was not, however, distrib-uted uniformly among the 346 western non-metropolitan counties. In 1982 in 119 of thesecounties (34.4%), there was no employment inhigh tech sectors, and only 110 of the non-metro counties (31.8%) reported more than 50persons employed in high tech manufacturing.These percentages suggest that western non-metro high tech employment is relatively con-centrated, and thus the economic development

'Throughout this article, "metropolitan" and "urban" refer toMetropolitan Statistical Areas (MSAs). "Rural" and "nonmetro-politan" are used interchangeably to refer to nonMetropolitan Sta-tistical Areas (nonMSAs).

Western Journal of Agricultural Economics, 16(2): 331-344Copyright 1991 Western Agricultural Economics Association

Western Journal of Agricultural Economics

Table 1. High Technology ManufacturingIndustries

StandardIndustrial

Codea Industry Group

281 Industrial inorganic chemicals282 Plastic materials, synthetics283 Drugs286 Industrial organic chemicals289 Miscellaneous chemical products291 Petroleum refining348 Ordnance and accessories, n.e.c.351 Engines and turbines353 Construction and related machinery356 General industrial machinery358 Office and computing machines362 Electrical industrial apparatus365 Radio and TV receiving equipment366 Communication equipment367 Electronic components, accessories372 Aircraft and parts376 Guided missiles, space vehicles381 Engineering, scientific instruments382 Measuring and control devices383 Optical instruments and lenses384 Medical instruments and supplies385 Ophthalmic goods386 Photographic equipment and supplies387 Watches and clocks

a 1977 Standard Industrial Classifications from Armington, Harris,and Odle.

benefits associated with the decentralization ofthese firms are spatially limited.

The purpose of this study is to determinethe distinguishing characteristics of westernnonmetropolitan counties in which high techmanufacturing has located. Insight into factorsassociated with high tech plant locations andemployment should enable rural communitiesto (a) better assess their potential as locationsfor high tech firms and (b) identify programareas which may augment the communities'competitive advantages in attracting or gen-erating high tech activity. The article is orga-nized as follows. First, high technology man-ufacturers are identified and the characteristicsand locations of western nonmetro high techfirms are summarized. For this study, the Westis defined as the 11 contiguous states in theMountain and Pacific census divisions (Ari-zona, California, Colorado, Idaho, Montana,Nevada, New Mexico, Oregon, Utah, Wash-ington, and Wyoming). Second, an economet-ric analysis of county-level data is undertakento determine if the existence and magnitude

of high tech employment are associated sig-nificantly with selected nonmetropolitancounty characteristics. The Tobit estimationprocedure was used to estimate the relation-ships between high tech employment andcounty characteristics from the censored data.The nonmetropolitan high tech employmentdata were disaggregated by plant ownershiptype (branch plants vs. unit plants) to deter-mine if community factors associated withplant location varied by locus of control. Fi-nally, the findings are summarized and policyimplications are suggested.

Overview of Western High TechManufacturing

High Tech Manufacturing

Definitions of high technology industries vary;however, they generally are based on the fol-lowing criteria (Etzioni and Jargowsky):

(a) a high percentage of an industry's grossrevenue or output is dedicated to researchand development,

(b) the industry's workforce includes a highpercentage of scientists and engineers, and/or

(c) a high percentage of laborers characterizedas highly skilled.

The definition used in this study is that de-veloped in 1983 by Armington, Harris, andOdle of the Brookings Institution. This defi-nition (based on national data) classified anindustry as high technology if (a) more than8% of its employees were in scientific, engi-neering, and technical occupations, and at least5% of industry employment was in the morenarrow class of scientific and engineering oc-cupations; or (b) expenditures for research anddevelopment were a relatively large percent(greater than 5%) of product sales. Twenty-four manufacturing industries were identifiedunder these criteria (table 1).

Employment Patterns

Metropolitan areas in the 11 contiguous west-ern states have been relatively successful inattracting and generating high technologymanufacturing employment (table 2). The1975-82 metropolitan high tech growth rateexceeded 50%, and total high tech employ-

332 December 1991

High Tech Location 333

Table 2. High Technology Manufacturing Employment in the Western States, 1975-82

Metropolitan Nonmetropolitan

State 1975 1982 % Change 1975 1982 % Change

Arizona 39,728 73,836 85.9 956 1,777 85.9California 538,498 782,101 45.2 2,029 4,175 105.8Colorado 30,727 65,532 113.3 1,142 1,596 39.8Idaho 189 1,647 771.4 1,545 2,807 81.7Montana 766 925 20.8 392 718 83.2Nevada 817 2,906 255.7 597 301 -49.6New Mexico 4,061 7,791 91.8 1,182 2,110 78.5Oregon 14,730 21,071 43.0 1,250 4,498 259.8Utah 14,274 26,218 83.7 2,881 5,605 94.6Washington 61,550 90,393 46.9 2,064 2,744 32.9Wyoming 829 1,285 55.0 447 1,279 186.1Total 706,169 1,073,705 52.0 14,485 27,610 90.6

Source: Enhanced County Business Patterns, 1975 and 1982. The Enhanced County Business Patterns is a data file created by theNational Planning Data Corporation (Ithaca, New York) by additional processing of the U.S. Department of Commerce County BusinessPattern Series, which involves estimating suppressed data.

ment in the region's urban areas was in excessof one million jobs in 1982. This employmentwas concentrated in California; however, Ar-izona, Colorado, and Washington also devel-oped significant employment in this sector by1982.

Nonmetropolitan counties in the West alsobenefited from the region's growth in hightechnology employment. All states except Ne-vada experienced rapid nonmetro employ-ment growth in these industries from 1975 to1982, and over 13,000 high tech jobs wereadded to western rural areas during this period.Nonmetropolitan areas in Oregon, Wyoming,California, Utah, Arizona, Montana, Idaho,and New Mexico experienced the greatest rel-ative gains in high technology employment, allexceeding 75%. Despite this recent success ofsome areas, high technology manufacturerswere not a major source of employment in thenonmetropolitan areas of most western states.Only 27,610 high tech jobs existed in the non-metro West in 1982, and 14,278 (52%) of thesejobs were in nonmetro counties in California,Oregon, and Utah. 2

2 Nonmetro areas in the West differed in the types of high techmanufacturing represented. Enhanced County Business Patternsemployment data indicate that in three of the 11 western states(New Mexico, Washington, Wyoming), nonmetropolitan high techemployment was largely the result of the significant presence ofthe petrochemical industries. Nonmetropolitan areas in California,Colorado, Idaho, Montana, and Nevada were similar in that elec-tronic components, accessories, and apparatus manufacturers wereimportant employers. Other employers of significance in these fivestates include the manufacturers of measuring and control devices,construction machinery, and communication equipment. Non-

Preliminary analysis of the employment dataindicates that western nonmetro high techmanufacturing was concentrated in the morepopulous counties and in counties adjacent tometropolitan areas (table 3). Specifically, only96 of the 346 nonmetro counties (28%) hadpopulations exceeding 25,000, yet almost 80%of the high tech employment was located inthese counties. At the other extreme, 41% ofthe nonmetro counties had populations lessthan 10,000, yet only 5.4% of the high techemployment was located there. Also, the 264nonadjacent counties (76%) and the 82 adja-cent counties (24%) each had approximately50% of the employment in this sector.

This proclivity for locating in large and ad-jacent counties held for both unit plants andthe branches of multiplant operations (table3). For example, 57.3% of the nonmetro hightech employment was in the branches of mul-tiplant firms. Of this branch employment, 50%was distributed among the counties which wereadjacent to metro areas (24% of the total),and 82% was located in the 96 counties withpopulations greater than 25,000 (28% of thetotal). The pattern for unit plants was onlyslightly less concentrated with 74.9% locatedin the larger counties (25,000 plus) and 47.8%in counties adjacent to metro areas.

metropolitan areas in Oregon, Arizona, and Utah have developedhigh tech sectors unlike other western states. Arizona's nonmetro-politan technical sector was relatively diversified with the medicalinstruments and plastics materials industries providing the greatestemployment. The office and computing machines industry dom-inated nonmetropolitan high tech employment in Oregon, whileguided missile and space vehicle industries were dominant in Utah.

Barkley and Keith


Table 3. Distribution of Nonmetro High Tech Employment by County Size and AdjacencyStatus, Western States, 1982

Unit Plants Branch Plants All Plants

Non- Non- Non-Population Adjacent adjacent Adjacent adjacent Adjacent adjacent All Counties

.......................................................................................................... % ..................................................................................................

Small .9a 1.4 .3 2.8 1.2 4.2 5.4

(0-10,000) (5.0)b (36.2) (5.0) (36.2) (5.0) (36.2) (41.2)

Medium 1.1 6.6 1.7 5.5 2.8 12.1 14.9(10,000-25,000) (5.9) (24.7) (5.9) (24.7) (5.9) (24.7) (30.6)

Large 18.4 14.4 27.5 19.5 45.9 33.9 79.8(25,000+) (13.2) (15.0) (13.2) (15.0) (13.2) (15.0) (28.2)

Total 20.4 22.4 29.5 27.8 49.9 50.2 100.1(24.1) (75.9) (24.1) (75.9) (24.1) (75.9) (100.0)

Source: Compiled from Enhanced County Business Patterns, 1982.a Percent of total western nonmetro high tech employment in the size-adjacency category.b Percent of the 340 western nonmetro counties in the size-adjacency category.

Community Characteristics and IndustrialLocation

Previous research on the locational determi-nants of manufacturers found that nonmetroplant locations were influenced by labor andindustrial site availability, access to markets,labor costs, and, at times, local taxes.3 Thesefindings are consistent with the product lifecycle and spatial division of labor theories(Vernon; Thompson; Suarez-Villa; Marku-sen). According to these theories, rural areasare viable locations for the slowly growing, low-profit, mature manufacturers. Such industriesgenerally have standardized production pro-cesses, and thus can significantly reduce pro-duction costs by locating in areas with low-cost labor and land.

Schmenner, Huber, and Cook demonstrat-ed, however, that the locational attributes con-sidered important to manufacturers vary sig-nificantly depending on product type, missionof the plant, and production process. Thus,community characteristics found to be impor-tant to mature manufacturers (labor and in-dustrial site availability, proximity to markets,community infrastructure) may not reflect lo-cational factors important to firms in the hightech sector. Nonmetropolitan high tech firmsgenerally are more skilled-labor intensive than

3Earlier studies of manufacturing locations in nonmetropolitanareas include: Dorf and Emerson; Cromley and Leinbach; Miller1980; Erickson; Luloff and Chittenden; Erickson and Leinbach;Fisher; McNamara, Kriesel, and Deaton; and Walker and Cal-zonetti.

manufacturers in other sectors (Barkley, Dahl-gran, and Smith). As a result, communitieswith abundant low-skill, low-cost labor maynot be attractive locations to high tech firms.Moreover, nonmetropolitan high tech manu-facturers may be less sensitive to transporta-tion costs and less dependent on local marketsthan other firms. Oakey, and Smith and Bark-ley found that nonlocal purchases are exten-sive in high tech industries. These nonlocalinputs often come from numerous locations(Hagey and Malecki). In addition, the inter-mediate manufactured inputs of many hightech firms such as electronic components man-ufacturers are relatively inexpensive to trans-port, and the value of the final product is highrelative to transportation costs. In this case, asGoode points out, factors that normally reflectproximity to input and output markets (e.g.,distance from MSA, proximity to interstatehighways and rail lines, local availability ofintermediate inputs) may be of limited im-portance to high tech manufacturers. In thefollowing sections, the characteristics of non-metropolitan high tech manufacturing loca-tions in the West are analyzed to determine ifthe locational factors associated with nonmet-ro high tech firms differ from those noted inearlier studies of rural plant locations.

Data

The manufacturing plant location decision isviewed by neoclassical location theory as

334 December 1 991


selecting the production function-locationcombination that maximizes the owner's prof-it or utility function (Blair and Premus). Thisperspective suggests that firms quantify andcompare the costs and benefits associated witha finite number of alternative locations. Lo-cational factors included in this comparisonare access to input and product markets; laborcosts, availability, and quality; industrial siteavailability and quality; external economies(urbanization and localization); and percep-tions of the quality of life. Thus, location the-ory suggests that the volume of high tech man-ufacturing employment in nonmetrocommunities is related to subsets of commu-nity characteristics. Specifically,

HTEP = f(LFi, LOC, EXECi, QOL),

where HTE' denotes high tech manufacturingemployment in county i, LF' is a vector oflabor force characteristics in county i, LOC isa vector of locational characteristics, EXECis a vector of variables representing availabil-ity of external economies, and QOL is a vectorof local quality of life attributes.

Employment Data

1982 nonmetropolitan county employmentdata for manufacturers in the high tech sectorwere available in the U.S. Establishment En-terprise Microdata file (USEEM). USEEM is aproprietary data set developed by the Brook-ings Institution from Dun and Bradstreet'sDUNS Market Identifier files. County-levelemployment is provided by plant ownershiptype (branch vs. unit plant) thus permitting acomparison of locational factors by locus ofcontrol. Unfortunately, disclosure laws pre-vented the acquisition of county-level em-ployment for specific high tech industries.However, the aggregated data should provideinsight into factors affecting location andgrowth.

County Characteristics

Measures for nonmetropolitan county char-acteristics were obtained from the MontanaState University City and County Data Base,a computerized collection of data from the Cityand County Data series published by the U.S.Department of Commerce. The independentvariables selected for this research were thosefactors found to be significant in earlier non-

metro industrial location studies plus charac-teristics suggested by case studies and "com-mon knowledge" (e.g., universities andclimate). County characteristics for 1970 wereselected to insure a reasonable lag between lo-cal attributes and the development of a hightech sector. Quite clearly, some simultaneitywould exist in those counties having significanthigh tech manufacturing employment in 1970.Much of the growth in high tech employmentis relatively recent; thus, this simultaneityshould not be a serious problem.4

In total, over 70 measures were chosen torepresent the counties' socio-, demographic,and economic environments (appendix, tableAl). Obviously, much redundancy and cor-relation exists among these characteristics. Asan aid to selecting uncorrelated explanatoryvariables, a factor analysis was completed forthe variables using the SAS Factor programwith standard parameters (Dorfand Emerson).Table 4 provides the factors and associatedvariables with their factor weightings. The re-sults indicate that the community character-istics clustered fairly "cleanly" into 11 prin-cipal categories. That is, differences among the346 western nonmetro counties may be rep-resented by 11 characteristic groupings or fac-tors. Specifically, factor 1 included countycharacteristics closely associated with countysize and urbanization (e.g., number of doctors,dentists, and hospital beds; existence of col-leges or universities; employment in specificindustrial sectors; and government expendi-tures per capita). Thus, county population(POP) was selected as the proxy variable for(a) the availability of a broad array of publicand private services and (b) labor availability.A positive relationship between POP and HTEis expected.

Note that it would have been desirable toinclude "university present" and "governmentexpenditures per capita" as separate explana-tory variables. The factor analysis demon-strates, however, that for the nonmetro Westthese variables are highly correlated withcounty population. To avoid estimation prob-lems inherent with correlated explanatoryvariables, POP was selected to represent the"basket of goods" available in different coun-

4 Over 50% of the West's nonmetro high tech employment hasdeveloped since 1975. Thus, the simultaneity problem should beminimized by using 1970 community characteristics to explain1982 employment.

Barkley and Keith


Table 4. Factors and Loading Variables

Factor 1 Factor 2 Factor 3 Factor 4

rrnIX. " UrIDUfT AP 70 -TAA -TT -- 0SPOP .99 MD iC'YK ./Y rruvii.fD . .

Fem Emp .99 % Hi Sch .85 Farm Sales .64 Rural Frm Pop .39

NW Pop .98 Per Cap Inc .71 Ag & For Emp .74 Jul Cool .35

Tot W Pop .90 Med Fam Inc .6 FRMOWN .38

Universities .89 White Pov -. 51Val Fam Hs .87 Prop Tax .71Dentst .80 Med Gross Rent .72Govt Expend .89 % Urban .54Hosp Beds .54 NWPov -.46Rural Nonfarm

Pop .84MFG .95

Factor 5 Factor 6 Factor 7 Factor 8

Fem Emp .92 MIN.47 % Urban .52 MIN .40

Male Emp .94 % 18 + .41 NETMIG.50Fern Unemp .91 Land Area .48 % 4 + .40Male Unemp .92 % Frm Land .46 Work Out .39

Farm Size .47

Factor 9 Factor 10 Factor 11

MILB-.41 %18+ .49 ADJ-.42MILEM-.44 % 65 + .59

Note: Variable definitions are provided in appendix table Al.

ty-size groups. Unfortunately, we cannot iso-late the relative importance of the items in thebasket.

Factor 2 is an interesting mix of education,income, and property value measures. Thisgrouping of characteristics was interpreted torepresent the economic "well-being" of thecounty. Median School Years (MDSCYR) waschosen as the proxy variable for this factor,and a positive relationship between MDSCYRand HTE should be evident if high tech firmsseek locations with abundant skilled labor orstrong economies. Again, we would have pre-ferred to enter "median school years," "percapita income," and "mean property values"as proxies for "stock of human capital," "laborcosts," and "land costs," respectively. As not-ed above, the factor analysis demonstrates thatthis was not practical given the highly corre-lated nature of the variables.

Factor 3 included variables associated withfarm employment and sales. This factor shouldrepresent the importance of agriculture in thelocal economy. Because of high correlationsbetween total farm employment and total em-

ployment in other sectors, agricultural em-ployment [the sum of farm laborers and fore-men (FRMLAB) and farmers and farmmanagers (FRMOWN)] was divided by totalemployment to get agricultural employment asa percentage of total employment (AGEMP),which was then used as a proxy variable forthis factor. Agricultural counties may providea surplus labor pool. On the other hand, areasin the West which are dominated by agricul-tural employment appear to have relativelyfewer recreational opportunities and naturalamenities than counties less suitable for agri-culture. In addition, the agricultural sector usesfew high tech inputs with the possible excep-tion of chemicals. Therefore, there is no a prioriexpectation about the sign of this variable.

Factor 4 was the climatic factor. Januaryheating degree days (JANHT) (indicating theamount of heating required in the month ofJanuary) was selected as the proxy variable forclimate. While this variable does not neces-sarily reflect economic theory, a test of thepopular perception of high tech industry mov-ing to the "Sun Belt" was provided by this

336 December 1991


measure. No sign for the coefficient of JANHTwas hypothesized a priori.5

Factor 5 included total employment and un-employment by sex (Fer Emp, Fer Unemp,Male Emp, Male Unemp). The rate of un-employment (UNEMP) was selected to rep-resent this grouping, again because the corre-lation matrix indicated a very high correlationof total employment and unemployment topopulation. The unemployment rate was cal-culated by dividing the sum of male and femaleunemployment by the sum of male and femaleemployment and unemployment in a givencounty. This "rate" variable represents the rel-ative "tightness" of the labor market in a coun-ty, and a positive relationship between UNEMPand HTE should exist if high tech firms areattracted to areas with surplus labor.

Factor 8 related to total mining employ-ment. Again, because this employment wasclosely related to total population, mining em-ployment (MIN) was divided by total em-ployment to obtain a ratio (MINEMP) whichwas used in the analysis. This variable repre-sents the impact of petroleum and other min-eral and nonmineral sectors on high tech lo-cations and employment. Mining industries aremajor markets for high tech products; there-fore, it was hypothesized that the MINEMPvariable and HTE would be positively corre-lated.

Factor 11 identified counties adjacent tometro areas. A dummy variable for this ad-jacency was included (ADJ). Since metro areasprovide both markets for and spill overs fromhigh tech industries, a positive correlation be-tween ADJ and HTE was anticipated.

Finally,6 factors 6 and 7 included relativelyheterogeneous collections of community char-acteristics and not very strong factor loadings.Moreover, the measures for agricultural andmining activity in factor 6 and urbanization,commuting, and education in factor 7 wereaccounted for in other factors. Thus, the only

5 Metropolitan centers of high tech activity exist in both the"warmer" climates (Phoenix, Tucson, Los Angeles, Albuquerque)and the "cooler" climates (San Francisco, Portland, Seattle, Den-ver) of the West. Thus, neither the "northern" nor "southern"nonmetropolitan areas appears to be isolated from high tech mar-kets.

6 Factors 9 and 10 related to the existence or size of militaryactivity (MILB and MILEM) and the percentage of populationover 18 and over 65 (% 18+, % 65+), respectively. In earlier testsof various models, these variables were highly correlated with otherincluded variables and never significantly related to HTE and,therefore, were not used in the final model.

variable selected from these two characteristicgroupings was 1960 to 1970 net migration(NETMIG). Net migration is defined as in-migration minus out-migration and the differ-ence could be positive, zero, or negative. Thisvariable reflects the past growth history of thecounty and, indirectly, the general attractive-ness of the area to outsiders. A positive rela-tionship is hypothesized between HTE andNETMIG.

In summary, the factor analysis identifiedeight distinct county characteristics groupingswhich were considered relevant to firm loca-tion decisions. The characteristics identifiedwere very similar to those selected in earlierlocation studies. Specifically, the availabilityof public and private services, the availabilityof labor, presence of a university, and levelsof government expenditures have been shownto influence firm location decisions. In thenonmetro West, these characteristics werehighly correlated with county population andthus represented by one variable (POP). Coun-ty income, education, and property values arealso anticipated to affect location decisions.Again, however, these measures were highlycorrelated, and thus, must be represented byone measure (MDSCYR).

Labor force characteristics identified wereunemployment rate (UNEMP), agriculturalemployment as a percent of total employment(AGEMP), and mining employment as a per-cent of total (MINEMP). To these was addedmanufacturing employment as a percent of to-tal (MFGEMP), calculated by dividing totalmanufacturing employment (MFG) by totalemployment. The MINEMP and MFGEMPvariables are proxies for proximity to potentialmarkets, and positive relationships betweenHTE and MFGEMP and MINEMP are an-ticipated. Bender et al. have shown that mostnonmetropolitan communities are highly spe-cialized, that is, readily characterized as farm-ing, mining, or manufacturing towns. The co-efficients of the MFGEMP, MINEMP andAGEMP variables should provide insight intothe attractiveness of specialized economies tohigh tech firms.

To the locational characteristics identifiedby factor analysis (ADJ and JANHT) we addeda dummy variable for adjacency to an inter-state highway (HWY). Past studies have founda positive relationship between interstate ac-cess and economic development. Finally, localquality of life is an umbrella concept for which

Barkley and Keith

Western Journal ofAgricultural Economics

no one variable can adequately account. Anumber of the characteristic groupings do,however, provide some insight into quality oflife. For example, POP reflects the availabilityof local services and government spending percapita, MDSCYR is positively correlated withfamily income and property values and neg-atively related to the local incidence of pov-erty, ADJ reflects proximity to metropolitangoods and services, and NETMIG is a proxyfor the attractiveness of the county to outsid-ers. 7

Estimation Procedure

The empirical models to be estimated are:

HTEu = f(POP, HWY, ADJ, MDSCYR, JANHT,UNEMP, NETMIG, MFGEMP, AGEMP,MINEMP)

and

HTEb = f(POP, HWY, ADJ, MDSCYR, JANHT,UNEMP, NETMIG, MFGEMP, AGEMP,MINEMP)

where HTEu denotes county high tech man-ufacturing employment in unit plants during1982 and HTEb denotes county high techmanufacturing employment in branch plantsduring 1982.

In many earlier location studies, ordinaryleast squares (or generalized least squares) ap-proaches were used to estimate the relation-ships between local characteristics and localemployment in a specific sector. Peddle haspointed out, however, that employment datamay be censored, in the sense that there maybe significant numbers of counties (or otherlocational identifications) in which no ob-served employment occurs. As Amemiya,among others, has noted, the use of a least-squares estimator for censored data results inestimators which are inconsistent. The prob-lem is a significant one in nonmetro high techlocation analysis, as indicated by the fact that119 of the 346 counties had no high tech em-ployment.

The statistical approach used to correct forcensored data estimation problems is the Tobitmodel. According to Peddle (p. 304), the Tobitprocedure recognizes the special nature of

7 Data on miles of seacoast, number of lakes, number of amuse-ment parks, and other variables for specific amenities were avail-able. However, none of these measures were found to be signifi-cantly correlated with location in estimations not reported herein.

threshold values of independent variables andmakes use of the information contained incounties with zero employment. The Tobitmodel analyzes first the difference between zeroand nonzero values and then differentiating onthe basis of explanatory variables, betweenvarying nonzero levels of employment.

Estimation Results

The Tobit estimation results are presented intable 5 for high tech branch plants and table6 for high tech unit plants. Also provided arethe means and standard deviations of the in-dependent variables. Note that the estimationsoftware which was used (LIMDEP as devel-oped by Green) eliminated all observationswith missing variables. As a result, the esti-mations were based on 318 rather than 346counties. Maximum likelihood estimationswere used in order to provide the most con-sistent and efficient estimators (although het-eroskedasticity may remain a problem).

Branch Plants

The results of the Tobit analysis indicate in-teresting differences between the locational de-terminants of high tech branch and unit plants.High tech branch plant employment was pos-itively related to county population, net mi-gration rates, and proximity to metropolitanareas. Thus, branch plants were attracted toareas with abundant and growing labor mar-kets, readily available public and private ser-vices, and access to the product and input mar-kets of metropolitan areas. In addition, branchplants were more likely to locate in the north-ern rather than southern areas of the West.

The labor market characteristics of countieswith high tech branches were somewhat un-expected. Branch plant employment was neg-atively related to the local concentrations ofagricultural and manufacturing employment,yet the coefficient for concentration of miningactivity was not significant. These findingswould seem to indicate that, with the possibleexception of firms supplying the mining in-dustries, branch plants tended to locate ineconomies not dominated by manufacturingor agriculture. More specifically, the negativecorrelation between "percent of employmentin manufacturing" and high tech employmentmay be interpreted as indicating that local

338 December 1991


Table 5. Determinants of High Tech Manufacturing Employment, Branch Plants, Nonmetro-politan West, 1982

Adjusted TobitEmployment

Exogenous Variables Mean Std. Dev. Tobit Estimate

Intercept -2,327.34 -597.74(-1.97)**

Population (POP) 20,752.0 23,666.0 0.017 0.0044(5.85)***

Adjacent to MSA (ADJ)a 0.289 0.454 465.01 119.43(3.03)***

Interstate Highway (HWY)a 0.384 0.487 -112.39 -28.866(-0.84)

Heating Degree Days in January 1,093.0 294.0 1.06 0.27(JANHT) (3.90)***

Median School Years 11.99 0.68 39.69 10.19(MDSCYR) (0.40)

Unemployment Rate (UNEMP) 0.108 0.111 -560.07 -143.84(0.86)

Net Migration Rate (NETMIG) 31.54 26.43 4.63 1.19(1.49)

Agricultural Employment as a 0.157 0.109 -2,457.94 -631.28Percent of Total (AGEMP) (-2.71)**

Manufacturing Employment as a 0.107 0.087 -2,002.47 -514.30Percent of Total (MFGEMP) (-2.18)***

Mining Employment as a 0.038 0.075 599.46 153.96Percent of Total (MINEMP) (0.67)

Sigma 790.50(13.17)***

R2 .16Number 318a

Note: Triple asterisks indicate significance at the .01 level, double asterisks indicate significance at the .05 level, and an asterisk indicatessignificance at the .10 level.a Binary variable.

manufacturing markets were not important tothese plants, or alternatively, competition forlabor in manufacturing communities discour-aged high tech branch plant locations. The re-luctance of high tech branch plants to locatein agricultural counties may reflect the isola-tion or low natural amenities (fewer moun-tains, lakes, canyons, etc.) of these areas rel-ative to other western locations. Finally, thecounty unemployment rate was inversely butnot significantly related to high tech branchemployment. Thus, high tech branches do notappear to be attracted to "slack" rural labormarkets as were many of the early nonmetromanufacturers.

The final variable of interest is median schoolyears, our principal proxy for local income lev-els and quality (stock) of human capital. Thecoefficient on this variable was positive but notsignificant at the .10 level. Thus, while the

branches were not attracted to economicallystagnant communities (as evidenced by the co-efficients on unemployment and net migrationrates), no strong association was found withcommunities with high incomes and educa-tional levels.8

Unit Plants

The locational determinants for locally ownedhigh tech plants were similar to those just dis-cussed for branch plants, but with some inter-esting exceptions (table 6). The unit plants alsopreferred locations in populous counties, ad-jacent to metro areas, in northern states, withno concentration in agriculture or manufac-

8 The significance of sigma in the regression results is of interest.Sigma reflects the importance of censoring in the data in estima-tion.

Barkley and Keith


Table 6. Determinants of High Tech Manufacturing Employment, Unit Plants, Nonmetro-politan West, 1982

Adjusted TobitEmployment

Exogenous Variables Mean Std. Dev. Tobit Estimate

Intercept -403.04 -142.36(-2.03)**

Population (POP) 20,752.0 23,666.0 0.0033 0.0012(6.56)***

Adjacent to MSA (ADJ)a 0.289 0.454 29.78 10.51(1.20)

Interstate Highway (HWY)a 0.384 0.487 -7.96 -2.81(-0.36)

Heating Degree Days in January 1,093.0 294.0 0.06 0.023(JANHT) (r1.5)

Median School Years 11.99 0.68 27.02 9.54(MDSCYR) (1.65)*

Unemployment Rate (UNEMP) 0.108 0.111 -93.46 -33.01(-0.85)

Net Migration Rate (NETMIG) 31.54 26.43 0.35 0.12(0.70)

Agricultural Employment as a 0.157 0.109 -515.21 -181.98Percent of Total (AGEMP) (-3.58)***

Manufacturing Employment as a 0.107 0.087 -356.94 -126.08Percent of Total (MFGEMP) (2.56)***

Mining Employment as a 0.038 0.075 -79.75 -28.17Percent of Total (MINEMP) (-0.52)

Sigma 150.84(17.11)***

R2 .07Number 318

Note: Triple asterisks indicate significance at the .01 level, double asterisks indicate significance at the .05 level, and an asterisk indicatessignificance at the .10 level.a Binary variable.

turing. Yet, the unit plants' preferences for ad-jacency to MSAs and the northern regions ofthe West were much weaker than those ob-served for branch plants. Most interesting,however, is that the variable representing localincome levels and stock of human capital (me-dian school years) is positively and signifi-cantly correlated with unit plant locations. Thisfinding supports the hypothesis that an edu-cated/skilled labor force is of greater impor-tance to unit plants than to high tech branches.The positive relationship between educationallevels and unit plant employment is consistentwith both the product life cycle theory andrecent studies in entrepreneurial development(Cooper). The unit plants exhibited character-istics of firms in the earlier stages of the prod-uct cycle (Smith and Barkley). These estab-lishments employed a relatively large numberof individuals in research and developmentand in precision production. Thus, a well-ed-

ucated labor force should be an advantage.Also, the founders of new high tech firms aregenerally well educated, and communities witha large pool of such individuals should expe-rience a greater likelihood of high tech relatedentrepreneurial activity.

Decomposition of Tobit Results

McDonald and Moffitt have shown that theTobit coefficients can provide additional in-sight with both economic and policy impli-cations. The McDonald-Moffitt techniqueprovides a fraction by which the Tobit coef-ficients can be decomposed into two effects.Part one of the decomposition represents theeffect of a change in an exogenous variable onthe probability of the dependent variable beingabove the limit (i.e., a high tech firm exists inthe county). The second part is the effect of achange in an exogenous variable on the de-

340 December 1991


pendent variable assuming the dependentvariable is already above the limit (i.e., em-ployment is greater than zero). The coefficientsprovided by the second part more correctlyrepresent the relationship between a change incommunity characteristics and a change inemployment for counties with high tech firms.The amount of the adjustment depends on theproportion of the sample that is not at thelimits, with higher proportions resulting in asmaller reduction of the coefficients. In thisstudy, 48.1% of the nonmetro counties hadhigh tech unit plants and 28.3% had high techbranch plants.

The results of applying the McDonald-Mof-fitt adjustment to the Tobit coefficients areprovided in tables 5 and 6 under the heading"Adjusted Tobit Employment Estimate." Thecoefficients listed under the "Adjusted" head-ing are the Tobit coefficients for the relation-ship between a change in the independent vari-able and the expected value of employmentfor those observations for which employmentis not zero. The adjusted coefficients indicatethat levels of high tech employment (amongthose counties with high tech firms) are muchless closely related to changes in county char-acteristics than indicated by the aggregate To-bit results. Specifically, the adjusted coeffi-cients for the branch plant model wereapproximately one-fourth the magnitude of theaggregate Tobit estimates and the adjusted unitplant coefficients were only a little over one-third the value of the original estimates. Thesefindings may be interpreted to show that theselected county characteristics were better atdifferentiating between communities with andwithout high tech firms than for predicting em-ployment differences among communities withhigh tech employers. Or, alternatively, changesin select community characteristics will havea much larger impact on the probability ofattracting a high tech firm to a communitycurrently without any such plants than on thepossibility of increasing employment in a townwhich already has a high tech sector. 9

9 A potential shortcoming of the Tobit estimation procedure andthe McDonald-Moffitt adjustment is the underlying assumptionthat the same set of factors has the same influence on the existenceof high tech firms and on employment levels in these firms amongcounties (Norris and Batie). This may not be the case. Heckman(1976, 1979) offers an alternative procedure for dealing with cen-sored data which would allow for the examination of the selectioneffect (in this case, the choice of a location) independent from thesize of the activity (in this case, high tech employment). The Heck-

Conclusions and Implications

The locational preferences of nonmetropolitanhigh tech manufacturers are fairly straightfor-ward. In this study, high tech branch plantspreferred populous, rapidly growing countiesand locations near metropolitan areas. Coun-ties with relatively large manufacturing or ag-ricultural employment were avoided, possiblyto reduce competition for labor. Interestinglyenough, counties with high unemploymentrates also were avoided. These findings are notrepresentative of branches of mature manu-facturers seeking labor surplus locations.

The characteristics of counties with high techunit plants are less clear, perhaps due to therandomness of entrepreneurial activity. Pop-ulous counties with a well-educated labor forcewere relatively successful in attracting or gen-erating unit plant employment. These loca-tional attributes fit more closely the "commonknowledge" perceptions of the type of non-metropolitan area that might foster high tech-nology employment.

The findings of this study lead to severalimplications for community industrializationefforts. First, high tech employment in thenonmetro West is relatively concentrated, andthe counties which have been most successfulin attracting this employment are generally thelargest and most prosperous. Thus, the sparse-ly populated, isolated, low-amenity rural areaswill benefit little from development programstargeted at high tech manufacturers. Second,high tech firms prefer locations with diversi-fied economic bases. However, a communitywithout a history as a manufacturing centermay not be at a disadvantage in seeking hightech firms since, in this study, the firms avoid-ed locations with high concentrations of man-

man approach is a two-equation procedure involving estimationof a probit model of selection decision for all observations, cal-culation of a variable representing the sample selection bias effect(Inverse Mills Ratio), and incorporation of the variable repre-senting the selection effect into the estimation for those observa-tions where employment is positive.

The results of the Heckman estimation procedure for branchand unit plants are provided in table A2. The coefficients are notdirectly comparable to the Tobit results. However, with respect tothe independent variables which are significant, the results of thetwo models are very similar. For unit plants, the community char-acteristics significantly associated with the existence of a high techsector were also significantly related to employment levels. Forbranch plants, however, only county population was significantlyassociated with the existence of branch plant activity. The secondstage of the Heckman procedure for branch plants was very similarto the Tobit results. Thus the Heckman procedure indicates thatfor branch plants the assumption of similar sets of influencingfactors may not be valid.

Barkley and Keith


ufacturing employment. Finally, the high techmanufacturers may be locating in rural areasin an effort to reduce production costs (relativeto metro counties), but there is no indicationthat the lowest-cost rural locations are select-ed. Indeed, the high tech manufacturers (es-pecially the unit plants) are locating primarilyin communities with high median incomes andlow unemployment rates. Thus, programs ini-tiated to improve local services, educationalattainment, and quality of life could enhancethese communities' comparative advantage inthe competition for high tech employment.

[Received September 1989; final revisionreceived June 1991.]

References

Amemiya, T. Advanced Econometrics. Cambridge MA:Harvard University Press, 1985.

Armington, C., C. Harris, and M. Odle. "Formation andGrowth in High Technology Business: A RegionalAssessment." Washington DC: The Brookings Insti-tution, 1983.

Barkley, D. L. "The Decentralization of High-TechnologyManufacturing to Nonmetropolitan Areas." Growthand Change 19(Winter 1988):13-30.

Barkley, D. L., R. A. Dahlgran, and S. M. Smith. "High-Technology Manufacturing in the NonmetropolitanWest: Gold or Just Glitter." Amer. J. Agr. Econ.70(1988):560-71.

Barkley, D. L., J. E. Keith, and S. M. Smith. "The Po-tential for High Technology Manufacturing in Non-metropolitan Communities." WREP 97, Western Ru-ral Development Center, Corvallis OR, 1989.

Bender, L. D., B. L. Green, T. F. Hady, J. A. Kuehn, M.D. Nelson, L. B. Parkinson, and P. Ross. "The Di-verse Social and Economic Structure of Nonmetro-politan America." U.S. Department of Agriculture,Economic Research Service Report #49. WashingtonDC, 1985.

Blair, J. P., and R. Premus. "Major Features in IndustrialLocation: A Review." Econ. Develop. Quart. 1(1987):72-84.

Cooper, A. C. "Entrepreneurship and High Technology."In The Art and Science of Entrepreneurship, eds., D.L. Sexton and R. W. Smilor. Cambridge MA: Ballin-ger Publishing Co., 1989.

Cromley, R. C., and T. R. Leinbach. "The Pattern andImpact of the Filter Down Process in Nonmetropol-itan Kentucky." Econ. Geography 57(1981):208-24.

Dorf, R. J., and M. J. Emerson. "Determinant of Man-ufacturing Plant Location for Nonmetropolitan Com-munities in the West North Central Region of theU.S." J. Rgnl. Sci. 18(1978):109-20.

Erickson, R. A. "The Filtering Down Process: Industrial

Location in a Nonmetropolitan Area." Prof Geog-rapher 26(1976):254-60.

Erickson, R. A., and T. R. Leinbach. "Characteristics ofBranch Plants Attracted to Nonmetropolitan Areas."In Non-Metropolitan Industrialization, eds., R. E.Lonsdale and H. L. Seyler. New York: Halstead Press,1979.

Etzioni, A., and R. Jargowsky. "The Two Track Society:High Technology Basic Industry and the Future ofthe American Economy." Paper presented at the Cen-ter for Policy Research, Conference for a GrowingAmerica, March 1984.

Fisher, J. S. "Manufacturing Additions in Georgia: Met-ropolitan and Nonmetropolitan Differences from 1961to 1975." Growth and Change 10(April 1979):6-16.

Glasmeier, A. "Bypassing America's Outlands: RuralAmerica and High Technology." Report to Aspen In-stitute, Ford Foundation, Washington DC, 1988.

Goode, F. M. "Industrial Location Models: The Efficacyof More Refined Demand Variables." Growth andChange 17(January 1986):66-75.

Green, W. H. LIMDEP Version 5. PC Program and Man-ual. New York, 1988.

Hagey, M. J., and E. J. Malecki. "Linkages in High Tech-nology Industries: A Florida Case Study." Environ.and Planning 18(1986): 1477-98.

Heckman, J. J. "Sample Selection Bias as SpecificationError." Econometrica 47(1979):153-61.

"The Common Structure of Statistical Models ofTruncation, Sample Selection, and Limited Depen-dent Variables." Ann. Econ. and Soc. Measure.5(1976):475-92.

Luloff, A. E., and W. H. Chittenden. "Rural Industrial-ization: A Logit Analysis." Rural Sociology 49(1984):67-88.

Markusen, A. R. Profit Cycles, Oligopoly, and RegionalDevelopment. Cambridge MA: The M.I.T. Press, 1985.

Markusen, A. R., P. Hall, and A. Glasmeier. High TechAmerica. Boston MA: Allen and Unwin, Inc., 1986.

McDonald, J. F., and R. A. Moffitt. "The Use of TobitAnalysis." Rev. Econ. and Statist. 62(1980):318-21.

McNamara, K. T., W. R. Kriesel, and B. J. Deaton."Manufacturing Location: The Impact of HumanCapital Stocks and Flows." Rev. Rgnl. Stud. 18(1988):42-48.

Miller, J. P. "Non-metro Job Growth and LocationalChange in Manufacturing Firms." Rural Develop-ment Res. Rep. No. 24. U.S. Department of Agri-culture, Economic Statistics Cooperative Services,Economic Development Division. Washington DC,1980.

. "The Product Cycle and High Technology In-

dustry in Nonmetropolitan Areas, 1976-80." Rev.Rgnl. Stud. 19(1989):1-12.

Norris, P. E., and S. S. Batie. "Virginia Farmers' SoilConservation Decisions: An Application of TobitAnalysis." S. J. Agr. Econ. 19(July 1987):79-90.

Oakey, R. P. High Technology Small Firms: RegionalDevelopment in Britain and the United States. NewYork: St. Martin's Press, 1984.

Peddle, M. T. "The Appropriate Estimation of Intra-

342 December 1 991


metropolitan Firm Location Models: An EmpiricalNote." Land Econ. 63(1987):303-05.

Schmenner, R. W., J. C. Huber, and R. L. Cook. "Geo-graphic Differences and the Location of New Manu-facturing Facilities." J. Urban Econ. 21(1987):83-104.

Smith, S. M., and D. L. Barkley. "Labor Force Charac-teristics of'High-Tech' vs. 'Low-Tech' Manufacturingin Nonmetropolitan Counties in the West." J. Com-munity Develop. Soc. 19(1988):21-36.

Suarez-Villa, L. "The Manufacturing Process Cycle andthe Industrialization of the United States-MexicoBorderlands." Ann. Rgnl. Sci. 18(1984):1-23.

AppendixTable Al. Variable Names for Data fromMontana State University City and CountyData Base

Abbreviation Used inVariable Name Text/Tables

Adjacent to a MetropolitanStatistical Area

Average farm sizeCrime rateElevationEmployment in manufactur-

ing firms with 100+ em-ployees

Expenditure on educationGovernment expenditureLand area

ADJFarm Size

Govt ExpendLand Area

Male and female employment by job classification in:ClericalCraftsmenFarm laborers and

foremenFarmers and farm

managersLaborer (except farm)Managers and administra-

tors (except farm)Operators (except trans-

port)Private household workersProfessional and technicalService workersNot reported

FRMLAB

FRMOWM

Male and female employment by sectors (13 categories):Agriculture, forestry, and

fisheries Ag & For Emp

Business and repair servicesConstructionEntertainment and recrea-

tional servicesFinancial and real estateManufacturing

DurableNondurable

TotalMiningPersonal services

MFG

Thompson, W. R. "Internal and External Factors in theDevelopment of Urban Economics." In Issues in Ur-ban Economics, eds., H. S. Perloff and L. Wingor.Baltimore MD: Johns Hopkins University Press, 1968.

Vernon, R. "International Investment and InternationalTrade in the Product Cycle." Quart. J. Econ. 80(May1966):190-207.

Walker, R., and F. Calzonetti. "Searching for New Man-ufacturing Plant Locations: A Study of Location De-cisions in Central Appalachia." Rgnl. Stud.23(1989)15-20.

Table Al. Continued


Professional servicesPublic administrationTransportationWholesale and retail tradeNot reported

Median family incomeMedian gross rent for family

housingMedian school yearsMetro or nonmetroMigration from countyMiles of seacoast1980 value of family dwell-

ings1970 value of family dwell-

ingsNonwhite population below

125% of poverty levelNumber of amusement parksNumber of collegesNumber of commutersNumber of dentistsNumber of hospital bedsNumber of January heating

daysNumber of July cooling daysNumber of junior collegesNumber of medical schoolsNumber of military personnel

presentNumber of recreation lakesNumber of universitiesPer capita incomePercent graduated from high

schoolPercent of population over 18

years oldPercent of population over 65

years oldPercent 25 or older with four

or more years of collegePercent urbanPercent working outside

countyPercentage of land in farms

Med Fam Inc

Med Gross RentMDSCYR

NETMIG

Val Fam Hs

NW Pov

DentstHosp Beds

JANHTJul Cool

MILEM

UniversitiesPer Cap Inc

% Hi Sch

% 18 +

% 65 +

%4 +% Urban

Work Out% Frm Land

Barkley and Keith

Western Journal ofAgricultural Economics

AppendixTable Al. Continued


Population POPPopulation per square milePresence of a military base MILBPresence of an airlineProperty taxes Prop TaxProximity to an interstate

highway HWYRural farm population Rural Frm PopRural nonfarm population Rural Nonfarm PopTotal female employment Fer EmpTotal females unemployed Fem UnempTotal male employment Male EmpTotal males unemployed Male UnempTotal nonwhite population NWpopTotal white population Total W popValue of farm land per acreValue of farm sales Farm SalesWhite population below

125% of poverty level White Pov

Table A2. Determinants of High Tech Manufacturing Location and Employment, Unit andBranch Plants, Heckman Estimation Results

Branch Plants Unit Plants

Exogenous Variables Activity Presentb 1982 Employmentc Activity Presentb 1982 Employmentc

Intercept -2.54 -6,717.03 -3.08 -464.73(0.64) (2.27)** (-1.51) (1.36)

Population (POP) 0.00003 0.04 0.000026 0.0033(2.17)** (4.79)*** (3.29)*** (6.32)***

Adjacent to MSA (ADJ)a 0.65 1,127.40 0.26 42.84(1.37) (2.07)** (1.12) (1.12)

Interstate Highway (HWY)a -1.88 -287.33 -0.10 -10.69(-3.99)*** (-0.67) (-0.49) (-0.32)

Heating Degree Days in January 0.0013 2.52 0.0004 0.075(JANHT) (1.34) (2.83)*** (0.83) (1.07)

Median School Years 0.0045 116.30 0.22 32.21(MDSCYR) (0.14) (0.45) (1.26) (1.15)

Unemployment Rate (UNEMP) -0.70 -984.80 -0.73 -93.92(0.33) (-0.53) (-0.76) (-0.62)

Net Migration Rate (NETMIG) 0.0081 11.69 0.0046 0.620.72 (1.31) (0.98) (0.96)

Agricultural Employment as a -4.21 -6,727.29 -4.11 -699.10Percent of Total (AGEMP) (-0.94) (-2.51)** (-3.49)*** (-3.82)***

Manufacturing Employment as a -3.83 -4,968.31 -2.48 -401.66Percent of Total (MFGEMP) (-1.17) (- 1.63)* (-1.92)** (-2.01)**

Mining Employment as a 1.12 1,353.13 -0.97 -142.74Percent of Total (MINEMP) (0.50) (0.37) (-0.76) (-0.70)

Sigma 1,965.13 162.02(8.07)*** (6.31)***

R2 .21 .05 .27 .05Number 318 90 318 153

Note: Triple asterisks indicate significance at the .01 level, double asterisks indicate significance at the .05 level, and an asterisk indicatessignificance at the .10 level.a Binary variable.b The probit equation of the Heckman selection procedure.c The selection equation of the Heckman selection procedure (maximum likelihood estimate).

344 December 1991

the locational determinants of western nonmetro high tech manufacturers

Documents