~~—- 3-1713 October 1994 a LIBRARY TEXAS A&M UNIVERSITY JAN 18 T995 TEXAS STATE DOCUMENTS MEAT PRODUCT SELECTION: an Analysis for the away-from-Home and at-Home Markets TEXAS STATE DEPOSITORY Texas Agricultural Experiment Station Edward A. Hiler, Director The Texas A&M University System College Station, Texas [Blank Page in Origiad Bulletin} > Meat Product Selection: an Analysis for the away-from-Home and at-Home Markets Rodolfo M. N ayga, _]r., and Oral Capps, _]r.* *Respectively, Assistant Professor, Department of Agricultural Economics and Marketing, Rutgers University, and Professor, Department of Agricultural Economics, Texas A&M University. The authors acknowledge helpful comments from Carl Shafer, Ernie Davis, and Teo Ozuna. KEYWORDS: meat consumption, food away from home, food at home, socioeconomic and demographic fac- ‘tors, logit analysis. [Blank Page in Origafl Bulletin] A '5» Contents Executive Summary .................................................................................................................. .. iv Introduction ............................................................................................................................... .. 1 Literature Review ....................................................................................................................... .. 1 Conceptual Framework for the Analysis ................................... .............................................. .. 5 Data Source and Description .............................................. .................................................... .. 6 Empirical Results ....................................................................................................................... .. 8 Food-away-from-Home Logit Model .................................................................................... .. 8 Logit Models for Meats ........................................................................................................ .. 9 Conclusions and Areas for Further Research .......................................................................... .. 12 Literature Cited ........................................................................................................................ .. l3 Appendix: Maximum Likelihood Estimates of the Logit Models for Meats ............................ .. l5 iii Executive Summary Little is known about the demographic and socioeconomic characteristics of individuals either who eat away from home or who eat a particular meat product away from home or at home. We developed logit models, using the Individual Intake phase of the 1987-88 National Food Consumption Survey, to investigate the decision to eat food away from home (FAFH) and the decision to eat a particular meat product either away from home or at home. Results indicate that the following individuals are less likely to eat FAFH: individuals from the Northeast compared with those in the South; blacks and Hispanics compared with whites; unemployed individuals compared with employed individuals; food stamp recipients compared with nonrecipients; those on special diets compared with those not on special diets; and larger households compared with smaller households. The likelihood of eating FAFH is significantly affected by the age and income variables, ceteris paribus. The probability of eating FAFH is also significantly lower during the first and third quarters of the year compared with the second quarter of the year. Contrasting results are apparent between the FAFH and FAH logit models across the various meat products. Employed individuals, for instance, are more (less) likely to eat a particular meat product away from home (at home) than are unemployed individuals. This result could be re- lated to the fact that employed individuals might have less time to prepare home-cooked meals than do unemployed individuals. In addition, results generally indicate that the probability of eating a particular meat product away from home (at home) decreases (increases) as the house- hold size increases. This result implies a decreasing affinity to eating out as household size gets larger. The weekend variable is also a significant factor in most meat logit models for FAFH but not in logit models for FAH. This result could imply that during weekends, the likelihood of eating meat products away from home is greater than the likelihood of eating meat products at home. Blacks are generally more likely to eat poultry, fish, and shellfish but less likely to eat beef than are whites either away from home or at home. Individuals residing in central cities and suburban areas are also more likely to eat poultry, fish, and shellfish but less likely to eat beef and pork at home than are those residing in nonmetro areas. Males are more likely to eat beef and pork. Individuals on special diets are generally more likely to eat poultry, fish and shellfish, lamb, veal, and game but less likely to eat beef and pork than are those not on special diets. This information might have some important implications for the pork and beef industries. For instance, the beef and pork industries might have to continue emphasizing the “leanness” of their products in their promotion campaigns to recapture health- and nutrition-conscious con- sumers. The poultry, fish, and lamb industries, on the other hand, should not only continue their efforts in promoting the “healthfulness” of their products but also focus on attracting minorities (e.g., blacks) as well as individuals who live in nonmetro areas. The logit models in this study identified the types of individuals who are more likely to eat FAFH as well as the types of individuals who are more likely to eat various meat products either away from home or at home. The identification of these types of consumers is essential in analyzing consumption behavior and in developing specific marketing programs. iv Introduction Between the 1970-72 and 1988-89 periods, red meat consumption in the United States declined al- most 12 percent, while poultry and fish/shellfish con- sumption rose 7O and 3O percent, respectively. Spe- cifically, beef consumption declined about 16 percent; veal consumption declined by almost 40 percent; pork consumption fell by nearly 2 percent; and lamb con- sumption fell by almost 5O percent. On the other hand, chicken and turkey consumption increased 64 per- cent and 95 percent, respectively (Putnam, 1990). These consumption trends clearly indicate a shift in consumer demand in the U.S. toward poultry prod- ucts and away from red meat products. Factors that could have caused this consumption shift are changes in tastes and preferences, changes in relative prices of meat products, and changes in dietary and health standards. Organizations representing meat producer groups are increasingly conscious of health and nu- trition concerns in the promotion of products (Capps and Schmitz, 1991). In fact, the red meat industry has been fostering the development of meat products that are not only leaner and low in fat but also quick, easy, and convenient to prepare. ~ One of the most noticeable trends in consumer ’ food expenditure patterns in recent years is the grow- ing proportion of income spent on food away from home (FAFH). In fact, the percentage of disposable income going to FAFH has increased from 5.5 per- cent in 1970 to 6.2 percent in 1989. In contrast, the percentage of disposable income going to food at home (FAH) has declined from 10.8 percent in 1970 to 7.6 percent in 1989. These economic trends point to the increasing importance of FAFH consumption relative to FAH consumption. Changes in consumer demographics and lifestyles contribute to the increased popularity of FA FH. Some socioeconomic and demographic factors that come to mind are a growing number of women, married and single, in the work force; increasing need for the convenience provided by eating out; more families living on two incomes; the impact of advertising and promotion by large food service chains; and more people in the age group of 25 to 44 who are inclined to eat out often (Putnam and Van Dress, 1984). Only about 7 percent of all households now fit the stereo- typical family of a working husband, a wife who does not work for wages, and two children (Kinsey, 1990). Moreover, married couples with children are becom- ing fewer as a share of all households. The one-adult ~ households are the fastest growing and are likely to exhibit nonconventional food consumption patterns (i.e., FAFH consumption). Little is known about the demographic and socio- economic characteristics of individuals who either eat away from home or individuals who eat a particular meat product (i.e., beef, pork) away from home or at home. The restaurant and fast-food industries would benefit from a study providing information about the demographic and socioeconomic profiles of consum- ers who eat out. Likewise, the various meat industries (i.e., beef, pork, lamb, poultry, and fish) would also benefit from a similar study giving them information about the important influences on consumers’ deci- sions to eat a particular meat product either away from home or at home. This research attempts to fill these voids by using the Individual Intake phase of the 1987-88 National Food Consumption Survey. This study attempts to identify, in a definitive fashion, the demographic and socioeconomic characteristics of individuals who eat away from home and individuals who eat a particular meat product either away from home or at home. Literature Review Several studies have been conducted recently to examine consumer attitudes and preferences toward meat products. Many of these studies examined con- sumer attitudes and preferences toward a particular meat product. Some of these studies are also geared toward determining the effect of leanness on con- sumer demand (e.g., Branson et al., 1986; Skaggs et al., 1987; Menkhaus et al., 1988; Capps et al., 1988). Many in the meat industry, particularly the beef and pork industries, ascribe the recent downward shift in demand (i.e., beef, pork) to increased nutritional consciousness of consumers. Moreover, some indus- try executives also assert that consumer tastes and preferences changed away from beef and other red meats to white meats or toward less meat in general (Menkhaus et al., 1988). ' In particular, Branson et al. (1986) examined the effects of different degrees of meat leanness on con- sumer demand. Skaggs et al. (1987) analyzed the po- tential of marketing a branded, low-fat, fresh beef product. Their findings show that consumers are re- ceptive to beef products that are leaner and lower in fat. In 1988, Menkhaus et al., using logit regression technique, identified important factors that influence purchase and reorder decisions of a branded, low-fat, fresh beef product. The results from their study indi- cate the importance of health-related factors in influ- encing the decision to purchase leaner meats. Capps et al. (1988) identified several demographic and psychographic characteristics of consumers who buy lean meat products from a particular retail food chain in Houston. The analysis was performed using a probit model; the data source came from a telephone survey of 200 shoppers. Results of the survey indi- cate that consumers with the following characteris- tics are more likely to buy lean meat products: those who are 30 years of age compared with those 20 to 29 years of age; residents of Texas for more than 10 years compared with residents of Texas for less than 1O years; those who attended college compared with those who had not attended college; and fat-conscious consumers compared with nonfat-conscious consum- ers. Household size and the probability of buying lean meat products are positively related. There is, how- ever, no statistically significant relationship between the likelihood of buying lean meat products and the price consciousness of consumers. Several have also directed analyses toward identi- fying the sources of structural changes in U.S. meat consumption (e.g., Nyankori and Miller, 1982; Braschler, 1983; Chavas, 1983; Moschini and Meilke, 1984). Furthermore, other studies have focused on the effect of demographic factors such as family size and composition on meat consumption (e.g., Buse and Salathe, 1978; Cox et al., 1989; Blaylock and Smallwood, 1986; Lee, 1986). None of these studies, however, has examined the effect of socioeconomic and demographic factors on the decision to purchase various meat products on either the away-from-home or the at-home market. Few studies analyzing expenditure or consump- tion models for FAFH have been conducted recently. Previous studies of FAFH generally considered expen- ditures as a single category not disaggregating by type of facility or by type of food consumed (LeBovit, 1 967; Derrick et al., 1982; Prochaska and Schrimper, 1973; Kinsey, 1983; Redman, 1980; and Sexauer, 1979). Some of these studies on FAFH are descriptive in na- ture (LeBovit, 1967; Manchester, 1977). Derrick et al. (1982) examined the effects on the explanatory power as well as on the income elastic- ity estimates of including a broad range of demo- graphic variables of FAFH expenditure models under alternative measures of income. Using the 1972-73 Consumer Expenditure Survey, they found that delet- ing the demographic variables on the FAFH expendi- ture models overstated the income elasticity by about 25 to 35 percent. They then suggested that the exclu- sion of demographic variables in the analyses could lead to an erroneous classification of a commodity in terms of being a normal or luxury good. Overall, their results indicate the importance of incorporating de- mographic variables in the analysis of FAFH expendi- ture patterns. Prochaska and Schrimper (1973), 0n the other a hand, examined the effects of including a measure of \ the opportunity cost of homemaker’s time in a model for FAFH consumption. Using the 1965-66 NFCS data set from the U.S. Department of Agriculture, they found that the value of a homemaker’s time is an im- portant factor affecting food consumption, when the household is viewed as both a producing and a con- suming unit. Particularly, their results showed a posi- tive effect of opportunity cost of time on away-from- home consumption. Inclusion of the opportunity cost of time also resulted in lower estimates of income elas- ticity for FAFH than what would have been obtained in its absence. They also found that higher income households tend to buy higher priced meals and that expenditure elasticities are highest for urban house- holds and lowest for rural households. Presence of preschool-age children instead of adults tends to de- crease the number of meals eaten away from home. Moreover, blacks and other races tend to consume fewer meals away from home than do whites in the South and West but not in the Northeast and North- Central regions. Kinsey (1983) tested the effect of various sources of household income on the marginal propensity to consume FAFH for both white and nonwhite house- holds. Kinsey disaggregated the households by inten- sity of the wife’s participation in the labor force and by income. Using data from the Panel Study of Income Dynamics (PSID) and tobit regression analysis, she found that income earned by wives working away from home full time did not increase the marginal propensity to consume FAFH. Whites have higher expenditure and higher marginal propensities to con- sume but have lower income elasticities for FAFH than do nonwhites. The income elasticity for FAFH is less than one, indicating that these households do not consider FAFH a luxury. Kinsey (1983, p. 18), likewise, mentioned that “eating FAFH is not necessarily less time intensive than home-produced meals... and that those who eat out to reduce meal production time might frequent limited menu, family-type, or fast-food restaurants.” Redman (1980) added region, race, and residence as explanatory variables. This study examined the ef- fects of women’s time allocation and socioeconomic variables on the expenditures for meals away from home and for prepared foods. Applying the 1972-73 diary portions of the Bureau of Labor Statistics (BLS) and the 1973-74 Consumer Expenditure Surveys, Redman used expenditure rather than quantity data. Unlike the Prochaska and Schrimper (1973) study, age and educational level of the woman were included ‘~ explicitly in the model, along with race, region, and urbanization. The regression results indicate that household expenditures on foods requiring little preparation time are influenced by the women’s char- acteristics that affect the allocation of their time to household production. Family income has a positive effect, while family size has a negative effect on FAFH consumption. Black households spend significantly less on prepared foods than do nonblack households. Families with preschool children spend significantly less on FA FH than do those with older children. House- holds in urban areas spend significantly more on FAFH than do households in nonurban areas. Younger women eat out more often than do older women, who instead buy more prepared foods. Employed women also buy more prepared foods but not more away-from- home meals. This result substantiates the claim that employed household managers spend more on pre- pared foods than do unemployed household manag- ers. These results imply that the restaurant and fast- food industries should cater to young families with- out small children, while the prepared-food industries should orient toward older families in which the wife is in the labor force. Sexauer (1979) examined the effect of demo- graphic shifts and changes in the income distribution on FAFH expenditures. Sexauer demonstrated that specification bias will result if demographic and in- come distribution shift effects are significant and are ignored. In particular, Sexauer quantified the degree to which aggregate expenditure on FAFH depends on the distribution of income and of households among population subgroups. Using the 1960-61 Bu- reau of Labor Statistics and U.S. Department of Agri- culture Survey of Consumer Expenditure and Incomes and the 1972-73 Bureau of Labor Statistics Consumer Expenditure Survey, he analyzed the effects of popu- lation and income shifts between 1960-61 and 1973- 74 among 40 population subgroups on FAFH expen- diture. Using ordinary least squares and a linear func- tional form, the regression results indicate the impor- tance of sociodemographic factors. In particular, the results indicate that about 22.3 percent of the ob- served changes in average aggregate expenditure on FAFH might be explained by demographic shifts. Sexauer’s (1979) results also imply that married couples in households with and without wife in the labor force spend about the same amount of money on food but that the two-earner households substi- tute FAFH for food prepared at home. Lee and Brown (1986) investigated food consump- tion at home and away from home using the 1977-78 National Food Consumption Survey data set and a switching regression technique. They emphasized the household’s choice of whether or not to eat out and the factors affecting away-from-home and at-home food expenditures. The results suggest a positive re- lationship between income and eating away from home. However, once a household has decided to eat out, income has very little influence on how much is spent on food unless its income is more than $20,000 per year. Suburban household members are more likely to eat away from home than are those in rural areas or central cities. Moreover, the following house- holds are more likely to eat away from home: house- holds in the North-Central region; households with an employed female head; households with a female head who is more educated; households with both male and female heads; households with male mem- bers older than 4 years and younger than 26 years; and households with female members older than 4 and younger than 50 years. Lippert and Love (1986) used the BLS 1980 Di- ary Consumer Expenditure Survey to identify the direction and magnitude of the relationship be- tween expenditures for FAFH and prepared foods and socioeconomic characteristics of the house- holds between 1972-73 and 1980. They included only consumer units headed by married couples. The results indicate that in 1980, higher family in- come, wife’s employment and college education, a metropolitan residence, and larger family size were associated with higher expenditures for FA FH. Vari- ables having a greater negative impact in 1980 than in 1972-73 are the presence of children, a rural resi- dence, and being in the black race. S0 far, only McCracken and Brandt (1986) have examined FAFH expenditures by facility (restaurants, fast-food facilities, and other commercial facilities). Using the 1977-78 National Food Consumption Sur- vey data set and tobit analysis, they identified and measured the influence of factors affecting FAFH consumption by type of facility. The factors included in the analyses are education and age of the house- hold head, retirement status of household head, geo- graphic location of household (both regional location and degree of urbanization), race, household size and composition, and a variable indicating FAFH consump- g tion mostly on a weekday or a weekend. McCracken and Brandt (1986) also included a value of the household’s time, which was estimated using a sto- chastic-censoring model. The stochastic-censoring model consists of a potential market-earnings equa- tion, a reservation-earnings equation, and a sample selection rule that determines whether or not an in- dividual participates in the labor market. They found that age and retirement status of the household head, being nonwhite, and whether observations had been recorded during the week all had negative effects on total FAFH expenditure. Increases in income were associated with increases in FAFH expenditure, but at a decreasing rate. In addition, the value of the house- hold manager’s time was positively related to total FAFH expenditure, fast-food expenditures, and other commercial expenditures but was only marginally sig- nificant for restaurant expenditures. McCracken and Brandt (1986) suggested that these results could indi- cate that individuals eat at restaurants for reasons other than to save time and that eating away from home in fast-food restaurants depends less on income than on the value of food preparer’s time. Household- size and composition have stronger effects on spend- ing at fast-food and other commercial facilities than at restaurants. This study by McCracken and Brandt (1986), however, used data from the 1977-1978 Na- tional Food Consumption Survey and, therefore, may not reflect current market conditions. Table 1 summarizes the studies previously con- ducted on FAFH expenditure or consumption. Many of these studies have focused their analyses on the sociodemographic and economic factors affecting the away-from-home food consumption and expenditure using cross-sectional data from national samples. Com- mon sociodemographic factors considered were in- come, household size, urbanization, region, race, employment, and education. Some of the results from these studies differ in relative importance of these factors on FAFH consumption or expenditures, pri- marily resulting from the use of different consump- tion models, data bases, and estimation techniques. Drawing on household production theory, some of these studies incorporated a measure of the op- portunity cost of time in their models. Most studies, however, used different measures of the value of household time. Prochaska and Schrimper (1975), for instance, used estimated wage rates (based on “out of sample” information) of homemakers as a measure of opportunity cost of time. Redman (1980), on the other hand, used a dummy variable that distinguished between employed and unemployed married women. McCracken and Brandt (1986) employed a stochas- tic-censoring model to estimate the value of the household’s time. Most of these studies examined FAFH expendi- tures or consumption within the context of consumer market goods expenditure. No studies as yet have analyzed the effect of sociodemographic and eco- nomic factors on the decision to consume FA FH. Like- wise, none of these studies have determined the so- cioeconomic and demographic factors that affect the Table 1. Selected studies on food-away-from-home (FAFH) expenditure and consumption. Data set“ 1972-73 CES Hesearcher(s) Derrick et al. (1982) Prochaska and 1965-66 NFCS Schrimper (1973) (spring portion) Kinsey (1983) a PSID Redman (1980) 1972-73 and 1973-74 BLS, CES Sexauer (1979) 1960-61 and 1972-73 BLS, CES Lee and Brown (1986) 1977-78 NFCS Sociodemographic factors considered Income, household size, age, education, region, urbanization, employment, race, marital status Urbanization, income, race, region, employment, number of children Various sources of income, race, employment, household size Income, employment, family composition, education, age, region, race, urbanization Family size, age, urbanization, education, sex, employment, income Income, urbanization, region, employment, education, race, household size Focus of the study Impact of demographic variables on FAFH expenditure Effect of opportunity cost of homemaker’s time on FAFH consumption Effect of various sources of income on marginal propensity to consume FAFH Impact of socioeconomic factors and women's time allocation on FAFH Effects of demographic shifts and income distribution changes on FAFH expenditure Factors affecting away-from-home and at-home consumption and the decision to eat out McCracken and 1977-78 NFCS Education, age, retirement, region, race, Factors affecting FAFH expenditure by Brandt ( 1986) urbanization, income, household type of facility composition and size Lippert and Love (1986) 1980 BLS, CES Income, employment, education, family Relationship between FAFH expenditure size and composition, region, race, urbanization and socioeconomic characteristics of household between 1972-73 and 1980 ‘BLS, CES = Bureau of Labor Statistics, Consumer Expenditure Survey. NFCS = National Food Consumption Survey. PSID = Panel Study of Income Dynamics. decision to eat a particular meat product either away from home or at home. Conceptual Framework for the Analysis To investigate the decision to eat FAFH, we used logit analysis to estimate a model in which the likeli- hood of eating away from home is a function of a set of exogenous variables. The specific purpose of this analysis is to identify, in a definitive fashion, the de- mographic and socioeconomic characteristics of in- dividuals who have eaten away from home. Furthermore, we also used logit analysis in model estimation would investigate the decision to eat the following meat products: (1) beef; (2) pork; (3) lamb, veal, and game; (4) poultry; and (5) fish and shellfish either away from home, at home, or both. The logit analyses center on the hypothesis that a set of variables influence the decision to eat FAFH and the decision to eat various meats either away from home, at home, or both. The logit models are speci- fied as follows: PROB = b0 + blurbanl + b2urban2 + b3region1 + b4region2 + b5region4 + b6race2 + b7race3 + b8race4 + b9hisp1 + bwsexl + buempl-oyl + blzfstampl + budietl + bMhsize + blslogage + bwlogincome + bwweekend + blsquarterl + bl9quarter3 + b20quarter4 where PROB represents the following dependent vari- ables: (1) equal to 1 if the individual consumed FAFH, 0 otherwise; (2) equal to 1 if the individual consumed beef from FAFH, 0 otherwise; (5) equal to 1 if the individual consumed beef from FAH, 0 otherwise; (4) equal to 1 if the individual consumed beef from all foods, 0 otherwise; (5) equal to 1 if the individual consumed pork from FAFH, O otherwise; (6) equal to 1 if the individual consumed pork from FAH, 0 otherwise; (7) equal to 1 if the individual consumed pork from all foods, 0 otherwise; (8) equal to 1 if the individual consumed lamb, veal, and game from FAFH, 0 otherwise; (9) equal to 1 if the individual consumed lamb, veal, and game from FAH, O otherwise; ( 10) equal to 1 if the individual consumed lamb, veal, and game from all foods, O otherwise; (11) equal to 1 if the individual consumed poultry from FAFH, 0 otherwise; (12) equal to 1 if the individual consumed poultry from FAH, O otherwise; (13) equal to 1 if the individual consumed poultry from all foods, O otherwise; (14) equal to 1 if the individual consumed fish and shellfish from FAFH, 0 otherwise; (15) equal to 1 if the individual consumed fish and shellfish from FAH, 0 otherwise; (16) equal to 1 if the individual consumed fish and shellfish from all foods, 0 otherwise. The independent variables refer to the following: urbanl = 1 if individual resides in a central city, 0 otherwise; urban2 = 1 if individual resides in a suburban area, 0 otherwise; regionl = 1 if individual is in the Northeast, 0 other- wise; region2 = 1 if individual is in the Midwest, 0 other- wise; region4 = 1 if individual is in the West, 0 otherwise; race2 = 1 if individual is black, 0 otherwise; race5 = 1 if individual is Asian or Pacific Islander, 0 otherwise; race4 = 1 if individual is of some other race, 0 other- wise; hispl = 1 if individual is Hispanic, 0 otherwise; sexl = 1 if individual is male, O otherwise; employl = 1 if individual is employed, O otherwise; fstampl = 1 if individual is receiving food stamps, O otherwise; dietl = 1 if individual is on a special diet, O other- wise; hsize = household size; logage = the logarithm of age; logincome = the logarithm of income; weekend = 1 if the 3-day intake of the individual occurred mostly during a weekend, 0 otherwise; and quarterl , quarter}, and quarter4 correspond to a set of binary variables that measure seasonality, (quar- terl = 1 if January to March; quarter3 = 1 if July to September; quarter4 = 1 if October to December) (ref- erence category, April to June). One classification is eliminated from each group of variables for estimation purposes. The base group is composed of individuals who satisfy the following description: reside in a nonmetro area (urban3); re- side in the South (region3); are white (racel); are nonhispanic (hisp2); are female (sex2); are not em- ployed (employ2); are not participating in the food stamp program (fstamp2); are not on a special diet (diet2); and the 3-day intake occurred mostly during a weekday (weekday). Household income is used in- stead of individual income because the National Food Consumption Survey (N FCS) data set provides income information only for the household and not for an individual. The analyses on meats are separated into three different food sources: FAFH, FAH, and all foods eaten to determine whether different factors affect the likelihood of eating a particular meat product across these three food sources. Data Source and Description The data source used in this study is the Individual Intake phase of the 1987-88 National Food Consump- tion Survey (NFCS) of the U.S. Department of Agri- culture (USDA). This data set is the most recent of the national surveys of household food consumption conducted by USDA. Data collection for the 1987-88 NFCS data set started on April 1987 and continued through August 1988. The 1987-88 survey contains two parts: (1) household food use and (2) individual intake. The household phase provides information on food used by the household for a 1-week period and on the cost of that food. The Individual Intake phase, on the other hand, provides 3 days of information on food intake of household members. The Individual Intake phase of the 1987-88 NFCS data set marks only the fifth time that the USDA has collected nationwide information on the dietary intakes of individual house- hold members. The survey also provides various sociodemographic information on each household and household member. The 1987-88 NFCS sample was designed using a multistage, stratified, area-probability sampling method. According to the USDA, the stratification plan accounted for geographic location, degree of urban- ization, and socioeconomic considerations. This strati- fication process resulted in a total of 60 strata (17 central city, 28 suburban, and 15 nonmetro areas), which correspond to the geographic distribution, urbanization, and the density of the population within the conterminous United States. The selection of households for the sample in a particular area was based on a prelisted number of housing units in the area as well as on the estimates of occupancy rates. More details about the data collection process can be obtained from the U.S. Department of Agriculture. . The Individual Intake phase of the 1987-88 NFCS data set provides data on 3 consecutive days of food and nutrient intake by individuals of all ages surveyed in the 48 contiguous states. The first day’s data were recorded from the individual’s 24-hour dietary recall. The period for this 1-day recall was from midnight to 11:59 p.m. on the day preceding the’ interview. This collection was done during an in-home personal in- terview. The second and third days’ data were col- lected during a self-administered 2-day dietary record. Respondents were asked about the sources of each food eaten. Sources were food that was eaten at home, food brought into the home but later eaten away from home, and food that was never brought into the home. USDA considers food from the first two sources to be from the home food supply. Thus, this study consid- ers food from the first two sources to be food at home (FA H), and the third source to be food away from home (FAFH). The USDA study does provide infor- mation about where the FAFH was obtained (i.e., res- taurants, school, fast-food establishments, or some- one else’s home). The Individual Intake data set gives information about individuals on the following variables: urban- ization; region; race; sex; employment status; food stamp participation; Women, Infant and Children program participation; National School Lunch and National School Breakfast Programs participation; special diet information; household size; age; house- hold income; food sources; and foods consumed. The response rate by household in the survey was low, i.e., below 35 percent. This is lower than in pre- vious NFCS data sets. USDA indicated that a major reason for this occurrence was “heavy respondent burden” in terms of the amount of information asked from each respondent. Survey results may then be biased if respondents and nonrespondents have sys- tematically different behavior. In terms of the population characteristics in the March 1987 Current Population Survey, the unweighted sample represented the population fairly well. Nevertheless, Bethlehem (1988) showed how reweighting can reduce the potential for nonresponse bias. Consequently, even if the sample was designed to be self weighting, Human Nutri- tion Information Service (HNIS) and statisticians at Iowa State University created weights for the indi- viduals in the sample to match the characteristics of the sample and the population. Weights were constructed separately for each of three sex/age groups and for 1-day intakes and 3-day intakes. The three groups that HNIS used were men age 20 and \ ‘a older, women age 20 and older, and persons less than 20 years of age. More information on the weighting procedure is available from HNIS. The number 0f days that food intake information was available for an individual varied. Thus, for some individuals the information was provided for only a 2-day or 1-day period. Because of the different inter- view processes employed in each of the 3 days of in- take, combining all the individuals with 1 day, 2 days, and 3 days of completed intake will ignore intra-indi- vidual effects or variations. The 1-day dietary recall, for instance, has been criticized because it depends upon memory and may not capture information rep- resentative of an individual’s usual intake. The record method, on the other hand, has been recommended because food consumption can be recorded immedi- ately after ingestion. Respondent burden is increased, however, and intake information may be biased (Pao et al., 1985). Moreover, more than 8O percent of the sample completed 3 days of intake. For these reasons, only individuals who completed 3-day intakes were included in the analyses. Data are weighted to main- tain representativeness of the sample by adjusting for nonresponse. The weights for the 3-day intake are used for this purpose. The original data set contained 11,045 individual respondents. However, after dropping individuals having less than 3 days of completed intake and indi- viduals without relevant socioeconomic and demo- graphic information, the resulting data set contains 6,276 observations. Table 2 shows the descriptive statistics of the ex- ogenous variables used in the regression analyses. About 21 percent of the sample resides in central city areas; 49 percent in suburban areas; and 3O percent in nonmetro areas. Most of the individuals (35 per- cent) included in the sample are from the South. Eighty six percent are white; 96 percent are non-His- panic; 45 percent are male; 58 percent are employed; 95 percent are nonrecipients of the food stamp pro- gram; 14 percent are on special diets; and about 16 percent ate food mostly on a weekend during the 3- day survey period. Moreover, the average age of the individuals is about 43 years, and the average house- hold size is approximately three. Average household income is closeto $30,000. Sixty-eight percent of the individuals in the sample consumed FAFH. The descriptive statistics of the de- pendent variables used in the logit analyses for meats are presented in Table 3. About 56 percent of the to- tal number of individuals in the survey consumed beef from the all foods eaten group. However, only 15 per- Table 2. Descriptive statistics of the exogenous variables. Variable Mean Std. dev. Range Urbanization Central city 0.21 0.4044 0-1 Suburban area 0.49 0.5000 0-1 Nonmetro area“ 0.30 0.4567 0-1 Region Northeast 0.20 0.3997 0-1 Midwest 0.27 0.4452 0-1 Southa 0.35 0.4762 0-1 West 0.18 0.3843 0-1 Race White“ 0.86 0.3380 0-1 Black 0.10 0.2970 0-1 Asian/Pacific Islander 0.01 0.0906 0-1 Other race 0.03 0.1571 0-1 Origin Hispanic 0.04 0.1855 0-1 Non-Hispanica 0.96 0.1855 0-1 Sex Male 0.45 0.4968 0-1 Female“ 0.55 0.4968 0-1 Employment status Employed 0.58 0.4935 0-1 Unemployed“ 0.42 0.4935 0-1 Food stamp participation Recipient 0.05 0.2219 0-1 Nonrecipienta 0.95 0.2219 0-1 Special diet Yes 0.14 0.3495 0-1 No“ 0.86 0.3495 0-1 Week variable Weekend 0.16 0.3682 0-1 Weekdaya 0.84 0.3682 0-1 Seasons Quarter1 0.29 0.4554 0-1 Quarter2a 0.41 0.4899 0-1 Quarter3 0.14 0.3508 0-1 Quarter4 0.16 0.3689 O-1 Age 43.30 18.37 15-99 Household size 3.03 1.46 1-12 Income 29,621 .80 23,927.8 3-300,000 a Refers to the omitted category in the analysis. cent and 46 percent of the sample consumed beef from FAFH and FAH, respectively. Likewise, the pro- portion of individuals consuming pork from all foods eaten, FAFH, and FAH is 44 percent, 10 percent, and 37 percent, respectively. About 3 percent of the sample consumed lamb, veal, and game from all foods eaten; 2 percent from FAH; and only 0. 5 percent from FAFH. From all foods eaten, 46 percent and 27 per- cent of the sample consumed poultry and fish/shell- fish, respectively. About 36 percent and 20 percent of the sample consumed poultry and fish/shellfish Table 3. Number of nonzero observations and means of the dependent variables used in the logit analyses for meats. Dependent variable All foods FAFH FAH Beef 3489 (0.56)a 933 (0.15) 2889 (0.46) Pork 2778 (0.44) 641 (0.10) 2339 (0.37) Lamb, veal, and game 181 (0.03) 30 (0.005) 151 (0.02) Poultry 2876 (0.46) 804 (0.13) 2285 (0.36) Fish and shellfish 1721 (0.27) 561 (0.09) 1280 (0.20) aNumber in parenthesis refers to the mean of the variable. from FAH, respectively. The proportion of the sample who ate poultry from FAFH is l5 percent, and the proportion of the sample ate fish or shellfish away from home is only 9 percent. Empirical Results Several variables possibly influence the decision to eat FAFH or the decision to eat a particular meat product either away from home or at home: (1) ur- banization, (2) region, (3) race, (4) ethnicity/origin, (5) sex, (6) employment status, (7) food stamp pro- gram participation, (8) special diet, (9) household size, (10) log of age, (11) log of income, (12) weekend, and (l3) seasonality. This section separately presents the regression results for the FAFH logit model and the logit models for meats. Food-awayfrom-Home Logit Model The maximum likelihood estimates for the logit analysis for FAFH are exhibited in Table 4. From the statistically significant coefficients and the changes in probabilities in Table 4, the results indicate that individuals in the Northeast are less likely to eat FAFH than are individuals in the South. Although not statis- tically significant, the negative signs of the coefficients of region2 and region4 variables also indicate that individuals in the Midwest and West are less likely to eat FAFH than are individuals from the South. Lee and Brown (1986) found that households in the North- Central region are more likely to eat away from home. Blacks are less likely to eat FAFH than are whites. Compared with whites, Asians/Pacific Islanders and individuals of other races do not to have a significantly different probability of eating FAFH. In terms of ori- gin, Hispanics are less likely to eat FAFH than are non- Hispanics. In particular, the probability of eating FAFH is significantly less by 0.1 560 for Hispanics relative to non-Hispanics. No statistically significant difference is found between the likelihood that males eat FAFH and the likelihood that females eat FAFH. As expected, Table 4. Maximum likelihood estimates of the logit model for FAFH. Changes in Variable Estimate Std. error probabilitya Intercept 1.35241 * 0.48814 0.28238 Urban1 0.09128 0.08514 0.01906 Urban2 0.03556 0.07087 0.00742 Fiegioni -0.18125" 0.08553 -0.03784 Region2 -0.03235 0.07910 -0.00675 Region4 -0.09852 0.08735 -0.02057 Race2 -0.32241* 0.10495 -0.06731 Fiace3 -0.35714 0.30711 -0.07457 Race4 -0.15635 0.19791 -0.03264 Hisp1 -0.74743* 0.16782 -0.15606 Sex1 0.03817 0.06072 0.00796 Employi 0.82299* 0.06346 0.17184 Fstampi -0.22910* 0.13346 -0.04783 Diet1 -0.10914 0.08273 -0.02279 Hsize -0.21611* 0.02298 -0.04512 Logage -1.05929* 0.07636 -0.00511° Logincome 0.37672’ 0.03957 0.0000027° Weekend 0.37916* 0.08575 0.07917 Quarteri -0.28839* 0.07144 -0.06022 Quarter3 -0.26608* 0.08895 -0.05555 Quarter4 -0.13477 0.08702 -0.02814 McFadden Fi-square 0.1150 R statistic 0.3320 Likelihood ratio test 907.56 Number of iterations 6 Ratio“ 0.6781 Percentage of right predictions = 72.5 ‘Statistical significance at the 0.05 level. “Equal to the product of the parameter estimates times the value of the logistic density function [B*f(z)]. At the sample means, the value of this density function (f(z)) is 0.2088, while the value of z is 0.8615. b Ratio of nonzero observations to the total number of observations. Note: The Fi statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGI Supple- mental Guide, SAS (1983) for further details. °Computed as [l3*f(z)/age] and [B*f(z)/income]. employed individuals are more likely to eat FAF H than are unemployed individuals. Assuming that employed individuals have a higher opportunity cost of time than do unemployed individuals, this result implies that the higher the opportunity cost of the individual’s time, the more likely he or she will eat away from home. Individuals are also more likely to eat FAFH during weekends than weekdays. As expected, food stamp recipients are less likely to eat FAFH than are nonrecipients. Although not sta- tistically significant, individuals on special diets are also less likely to eat FAFH than are individuals not on special diets. As expected, household size is negatively related to the probability of eating FAFH. Larger house- \ holds are less likely to eat FAFH than are smaller house- holds. The likelihood of eating FAFH is a function of the logarithm of age and income. The likelihood of eating FAFH is also significantly affected by age and income variables. Lee and Brown (1986) also found a positive relationship between income and the likeli- hood of eating away from home on a household level. The quarterly dummy variables, included to capture seasonal effects, are all statistically significant except for quarter4. Specifically, the probability of eating FAFH is lower in the months ofjanuary to March and July to September relative to the months of April to June (base quarter). Although not statistically signifi- cant, the likelihood of eating FAFH is also lower in the months of October to December than in the months of April to June. The McFadden R-squared (goodness-of-fit mea- sure) value shown in Table 4 is 0.1150. This value is reasonable considering the type of data (survey of individuals) used in the analysis. Another measure of goodness of fit is the correct classification of individu- als as either consuming or not consuming FAFH on the basis of the regression results. With a 50-50 classi- fication scheme, about 75 percent of the individuals in the sample are correctly classified as either con- suming or not consuming FAFH. Likelihood ratio tests are performed to determine the significance of some groups of variables in the model. Table 5 contains the results of the tests per- formed. The likelihood ratio tests on the urbanization variables as well as the regional variables are not sta- tistically significant when chi-square statistic is ap- plied at the 0.05 level with two and three restrictions, respectively. On the other hand, the likelihood ratio tests on the race and quarterly seasonal variables are statistically significant. These results indicate that the race variables as a group and the seasonal variables as a group contribute significantly to the explanatory power of the logit model. Logit Models for Meats The decision to eat various meats from FAFH, FAH, and all foods is also examined using logit analysis. Two Table 5. Results of the likelihood ratio tests on the logit model for FAFH. " No. of Likelihood ratio Variables restrictions test (chi-square) 1. Urbanization 2 1.16 2. Region 3 5.00 3. Race 3 10.59’ 4. Season 3 19.12’ ‘Statistically significant at the 0.05 level. related questions are addressed in these models. The first concerns whether significant differences in the probabilities exist between individuals who are eat- ing the same meat product from different sources: FAFH, FAH, and all foods. The second question per- tains to whether significant differences in the prob- abilities exist between individuals who are eating dif- ferent meat products from the same source: FAFH, FAH, or all foods. Five meat products are analyzed: (1) beef; (2) pork; (3) lamb, veal, and game; (4) poul- try; and (5) fish and shellfish. Hence, a logit model is estimated for each of these meat products from a par- ticular source: FAFH, FAH, and all foods. Table 6 shows goodness-of-fit measures of the logit models. The McFadden R-squared values for the FAFH models range from 0.0260 (beef) to 0.0820 (lamb,veal, and game); for the FAH models, the range is 0.0195 (beef) to 0.0355 (poultry). The all-foods models have McFadden R-squared values in the range of 0.0171 (beef) to 0.0294 (fish and shellfish). As expected, these McFadden R-squared measures are relatively low but are reasonable for analyses of survey data. In terms of the percentage of right predictions, the models seem to have predicted well; more than 85 percent of the individuals were correctly classified as either consuming or not consuming a particular meat prod- uct away from home. In addition, between 57 and 98 percent of the individuals are correctly classified as either consuming or not consuming a particular meat product at home (the same with all foods). Table 6. Goodness of fit of the logit models for meats McFadden Percentage of Model Fl-squared right predictions“ FAFH models Beef 0.0260 85.1 Pork 0.0310 89.8 Lamb, veal, game 0.0820 99.5 Poultry 0.0390 87.2 Fish, shellfish 0.0408 91.1 FAH models Beef 0.0195 57.1 Pork 0.0255 63.7 Lamb, veal, game 0.0219 97.6 Poultry 0.0355 65.5 Fish, shellfish 0.0284 79.8 All-foods models Beef 0.0171 57.3 Pork 0.0206 58.1 Lamb, veal, game 0.0206 97.1 Poultry 0.0265 59.2 Fish, shellfish 0.0294 72.8 aBased on a 50-50 classification scheme. The maximum likelihood estimates of the logit models for meats are exhibited in Tables A.1 to A. 15 in the Appendix. On the basis of statistically signifi- cant coefficients, the results from the beef logit mod- els (Tables A.1, A.6, and A.11 in the Appendix) indi- cate that individuals residing in central cities or sub- urban areas are significantly less likely to eat beef ei- ther from FAH and all foods but not from FAFH than are individuals residing in nonmetro areas. Individu- als from the Northeast and West, however, are less likely to eat beef from FAFH and all foods than are individuals from the South. The race variables as a group are significant factors in determining the like- lihood of eating beef from all three sources: FAFH, FA H, and all foods. In particular, whites are more likely to eat beef from FAFH, FAH, and all foods than are blacks and are more likely to eat beef from FAH than are Asians/Pacific Islanders. The seasonality variables, as a group, are significant factors in the logit models for beef from FAH and all foods but not from FAFH. In general, the likelihood of eating beef is higher (lower) in the third quarter (first quarter) from FAH and all foods than in the second quarter. Hispanics are more likely to eat beef from FAH and all foods than are non-Hispanics. Males, on the other hand, have a higher likelihood of eating beef from all three sources than do females. Employed in- dividuals are more likely to eat beef away from home but less likely to eat beef at home than are unemployed individuals. Food stamp recipients and individuals on special diets, on the other hand, are less likely to eat beef from FAH and all foods than are their counter- parts. As expected, household size is negatively re- lated to the probability of eating beef from FAFH but is positively related to the probability of eating beef from FAH and all foods. Generally, age is positively related to the likelihood of eating beef from all three sources except FAFH. Similarly, income is positively related to the likelihood of eating beef from FAFH and all foods. The weekend variable is positively related to the likelihood of eating beef away from home. The results from the logit models for pork (Tables A.2, A.7, and A.12 in the Appendix) show that indi- viduals residing in central cities or suburban areas are less likely to eat pork from all sources — FAFH, FAH, and all foods — than are individuals residing in nonmetro areas. Individuals from the Northeast and West, on the other hand, are less likely to eat pork from FAH and all foods than are individuals from the South. In contrast to the results describing beef in- take, whites are less likely to eat pork from FAH and all foods than are blacks and are less likely to eat pork from FAH than are Asians/Pacific Islanders. Seasonal- IO ity is not a significant factor in determining the likeli- hood of eating pork. Similar to the results describing beef intake, males are more likely to eat pork from FAFH, FAH, and all foods than are females. Likewise, employed individu- als are more likely to eat pork from-FAFH but less likely from FAH than are unemployedindividuals. In- dividuals on special diets are less likely to eat pork from either FAFH, FAH, or all foods than are those not on special diets. Household size is negatively re- lated to the likelihood of eating pork from FAFH but positively related to the likelihood of eating pork from FAH and all foods. Older individuals are more likely to eat pork from FAH and all foods than are younger individuals. Income is positive and significant in the FAFH model for pork but not in the FAH and all-foods models. The weekend variable is positively related to the likelihood of eating pork from FAH and all foods. The estimates from the logit models for lamb, veal, and game (Tables A5, A.8, and A13 in the Appen- dix) indicate that significant differences occur among the urbanization variables and among the regional variables from FAH and all foods. Specifically, indi- viduals residing in suburban areas are less likely to eat lamb, veal, and game than are individuals residing in nonmetro areas. Moreover, individuals in the North- east are more likely to eat lamb, veal, and game than are individuals in the South. Interestingly, results also show that food stamp recipients are more likely to eat lamb, veal, and game from FAFH, FAH, and all foods than are nonrecipients. On the other hand, in- dividuals on special diets are more likely to eat lamb, veal, and game from FAFH than are individuals not on special diets. Age is significantly positive in the FAFH model, and income is significantly positive in both the FAFH and all-foods models. Significant differences among regions exist in the probability of eating poultry in FAFH, FAH, and all foods (Tables A.4, A.9, and A. 14 in the Appendix). In fact, the likelihood ratio tests on the regional variables are significant in all three logit models for poultry. Specifically, individuals in the Northeast are less likely to eat poultry from FAFH and all foods than are indi- viduals in the South. Individuals in the Midwest and the West are also less likely to eat poultry from FAH and all foods than are individuals in the South. Indi- viduals in central cities and suburban areas are more likely to eat poultry from FAH and all foods than are individuals in nonmetro areas. As expected, blacks are more likely to eat poultry from FAFH, FAH, and all foods than are whites. Similarly, Asians/Pacific Is- landers and individuals of races other than white are more likely to eat poultry from FAH and all foods. \ E '\ The race variables as a group are statistically sig- nificant factors in the FAH and all-foods models, ac- cording to the likelihood ratio tests. Seasonal differ- ences are also evident in the results. In particular, poultry is less likely to be consumed from FAFH but more likely to be consumed from FAH during the first and third quarters of the year than during the second quarter. Hispanics have a higher probability of eating poul- try from FAH and all foods than do non-Hispanics. Likewise, individuals on special diets are more likely to eat poultry from FAH and all foods than are those not on special diets. Employed individuals, on the other hand, are more likely to eat poultry from FAFH but less likely from FAH than are unemployed indi- viduals. A negative relationship exists between house- hold size and the likelihood of eating poultry from FAFH. In contrast, a positive relationship exists be- tween household size and the likelihood of eating poultry from FA H. The coefficient of the age variable is significantly negative in the FAFH model but sig- nificantly positive in the FAH and all-foods models. A positive relationship also exists between the income variable and the probability of eating poultry from either FAFH or all foods. The weekend variable is positively related to the likelihood of eating poultry from FAFH. On the basis of likelihood ratio tests, the variables for urbanization, region, and race generally contrib- ute significantly to the explanatory power of the logit models for fish and shellfish (Tables A5, A.l0, and A. l5 in the Appendix). The results indicate that indi- viduals residing in central cities and suburban areas have higher probabilities of eating fish and shellfish ' from FAH and all foods than do individuals residing in nonmetro areas. In addition, individuals residing in central cities are more likely to eat fish and shell- fish from FAFH than are those residing in nonmetro areas. In terms of the regional differences, individu- als from the Northeast are more likely to eat fish and shellfish from FAH and all foods than are those from the South. Individuals from the West, however, are less likely to eat fish and shellfish from FAFH but more likely from FAH than are those from the South. The race variables as a group are statistically significant in the FAH and all-foods models but not in the FAFH model. Blacks and Asians/Pacific Islanders are more likely to eat fish and shellfish from FAH and all foods than are whites. Asians/Pacific Islanders are also more likely to eat fish and shellfish from FAFH than are whites. Individuals of other races, on the other hand, are less likely to eat fish and shellfish from FAH and all foods than are whites. ll Seasonal differences are also evident in the logit models for fish and shellfish. For instance, the prob- ability that an individual will eat fish or shellfish from FAH or all foods is lower during the third and fourth quarters of the year than during the second quarter of the year. Moreover, the likelihood that an individual will eat fish or shellfish from FAFH is lower in the fourth quarter than in the second quarter of the year Hispanics are more likely to eat fish and shellfish from FAH than are non-Hispanics. Employed individu- als are likewise more likely to eat fish and shellfish from FAFH than are unemployed individuals. As ex- pected, individuals on special diets are more likely to eat fish and shellfish from FAFH, FAH, and all foods than are those not on special diets. Household size is negatively related to the likelihood of eating fish and shellfish from FAFH and all foods. Age and income are positively related to the likelihood of eating fish and shellfish from all three sources. The weekend variable is positively related to the likelihood of eat- ing fish and shellfish from FAFH. Table 7 summarizes the logit models results con- cerning meats by showing the statistically significant estimates of the models. Generally, contrasting results are apparent between the FAFH and FAH models across the meat products. For example, employed individuals are generally more likely to eat a particu- lar meat product away from home than are unem- ployed individuals. Employed individuals, however, are generally less likely to eat a meat product at home than are unemployed individuals. Another example is the relationship between household size and the likelihood of eating a particular meat product. Gen- erally, the probability of eating a meat product away from home decreases as the household size increases. In contrast, the likelihood of eating a meat product at home increases as the household size increases. The weekend variable is also significant in most of the meat logit models for FAFH but not in the meat logit mod- els for FAH. Interestingly, individuals residing in cen- tral cities and suburban areas are more likely to eat poultry, fish, and shellfish but less likely to eat beef and pork from FAH or all foods than are those resid- ing in nonmetro areas. Blacks are generally more likely to eat poultry, fish, and shellfish but less likely to eat beef than are whites either away from home, at home, or both. Males are more likely to eat beef and pork than are females either away from home, at home, or both. The results also indicate that individuals on spe- cial diets are generally more likely to eat poultry, fish, shellfish, lamb, veal, and game but less likely to eat beef and pork than are those not on special diets. Table 7. Statistically significant parameter estimates of the logit models on meats. Variable Beef Pork All All Meat products Fish / Shellfish All Poultry All veal, and All foods FAFH FAH foods FAFH FAH foods FAFH FAH foods FAFH FAH foods FAFH FAH Likelihood ratio tests Urbanization Region Race Season Hispanic Male Employed Food stamp recipient On special diet Household size Age Income Weekend + + Note: An “s” indicates statistical significance at the 0.05 level. The (+) and (-) signs indicate the sign of the statistically significant parameter estimate. Conclusions and Areas for Further Research Logit models were developed to investigate the decision to eat FAFH and the decision to eat a par- ticular meat product either away from home or at home. These models provide a sociodemographic profile of individuals who are more likely to eat FAFH as well as of individuals who are more likely to eat various meat products either away from home or at home. The identification of these types of consumers is essential in analyzing consumption behavior and in developing specific marketing programs. Moreover, this information is useful for processors and produc- ers who want to anticipate future market changes and derived demands for their products. The results of these logit models should also be of interest to the restaurant and fast-food industries as well as the vari- ous meat industries (i.e., beef, pork, lamb, poultry, and fish). Various sociodemographic characteristics affect the likelihood of eating FAFH as well as the likelihood of eating a particular meat product either away from home or at home. Generally, the demographic and socioeconomic profiles of individuals eating the same 12 meat product are different across the three different food sources: FAFH, FAH, and all foods. Similarly, as mentioned in the “Executive Summary” section, con- trasting results are apparent between the FAFH and FAH logit models across the various meat products. Despite several studies on FAFH consumption, some areas still need further research. For instance, no studies have yet dealt with FAFH expenditures on a commodity basis (i.e., beef, pork, poultry, fish, etc.). Scant information is, therefore, available on demand parameters for FAFH expenditures by type of com- modity. This type of research could be handled with the use of the Consumer Reports on Eating Share Trends (CREST) data by the NPD Group. The CREST data series, collected by the NPD Group since 1976, is gathered via a comprehensive and detailed diary in which 12,800 U.S. households record their restaurant visits and purchases of meals, snacks, and beverages. The households, which are dispersed among the 48 contiguous United States, are recruited by mail using a stratified random quota sampling system. The CREST data series tracks more than 140 different food and beverage items. This series is the most comprehen- sive data set available on household purchase patterns of food in the FAFH market. l The effects of inventory demand and habits on consumer expenditure patterns on FAFH and FAH should also be examined. This research would, how- ever, require the availability of comprehensive time- series data with variable price and quantity data on food from the FAFH and FAH markets. The analysis in this type of study can be centered on the use of the Houthakker-Taylor state adjustment model. Generally, inventory demand tends to dominate habits in the short term. Likewise, short-run consumer behavior, as opposed to longer-run consumer behavior, is typi- cally influenced more by consumer inventories than by habits, particularly for food. It would be interest- ing to know the role of inventory demand and habits not only on aggregate FAFH and FAH but also on dis- aggregate commodities from FAFH and FAH markets. Structural change is related to the topic of habit formation. In demand analysis, shifts in the utility function may result from outside information and other external influences on the consumer or from variables related to past decisions (i.e., habit forma- tion). Investigating structural change in consumer behavior within the FAFH market would be worth- while. Ingco and Manderscheid (1988) discussed vari- ous ways to test for the presence of structural change in demand models. Although difficult, it would also be important to assess the magnitude of structural change, if any exist, in consumer behavior within the FAFH industry. Literature Cited Bethlehem, J.G. Reduction of Nonresponse Bias through Regression Estimation, journal 0f OjYic. Statistics, 4(l988):251-260. Blaylock, J., and D. Smallwood, U.S. Demand for Food: Household Expenditures, Demographics and Projec- tions, USDA, ERS, Technical Bulletin 1713, 1986. Branson, R., H. Cross, J. Savell, G. Smith, and R. Edwards, Marketing Implications from the National Consumer Beef Study, Western journal of Agricultural Econom- ics, ll(1986):82-91. Braschler, C., The Changing Demand Structure for Pork and Beef in the 1970s: Implications for the 1980s, Southern journal of Agricultural Economics, 15(1983):105-l10. Buse, R., and L. Salathe, Adult Equivalent Scales: An Alter- native Approach, American journal of Agricultural Economics, 60(l978):460-468. Capps, 0., Jr., D. Moen, and R. Branson, Consumer Char- acteristics Associated with the Selection of Lean Meat Products, Agribusiness, 4(l988):549-557. Capps, 0., Jr., andJ. Schmitz, A Recognition of Health and Nutrition Factors in Food Demand Analysis, Western journal of Agricultural Economics, 16(l99l):2l-35. 13 Chavas, J.P., Structural Change in the Demand for Meat, American journal of Agricultural Economics, 65(1983):148-153. Cox, T., R. Buse, and A. Alvarez, Effects of Demographics on Changes in At-Home Meat Consumption, The Eco- nomics of Meat Demand, R. Buse (ed.), Univ. of Wis- consin, Madison, 1986, pp. 217-24. Derrick, F., R. Dardis, and A. Lehfeld, The Impact of Demo- graphic Variables on Expenditures for Food Away from Home, journal of the Northeastern Agricultural Eco- nomics Council, 11(1982):1-10. Kinsey, J., Working Wives and the Marginal Propensity to Consume Food Away From Home, American journal of Agricultural Economics, 65( 1983): 10-19. , Diverse Demographics Drive the Food Industry, Choices, April-June 1990, p. 23. LeBovit, C., Expenditure for Food away from Home, Na- tional Food Situation, 152(1967):36-38. Lee, J.Y., Effect of Sociodemographics on At-Home Red Meat Consumption in the United States, The Economics of Meat Demand, R. Buse (ed.), Univ. of Wisconsin, Madi- son,l986, pp. 243-255. Lee, J., and M. Brown, Food Expenditures at Home and away from Home in the United States - A Switching Regres- sion Analysis, The Review of Economics and Statis- tics, 68(l986):l42-47. Lippert, A., and D. Love, Family Expenditures for Food away from Home and Prepared Foods, Family Economics Review, 3(1986):9-14. Maddala, G. S. Limited Dependent and Qualitative Vari- ables in Econometrics. Cambridge Univ. Press, 1983. Manchester, A.C., Eating Out, National Food Situation, 161(1977):25. McCracken, V.A., and J.A. Brandt. Analysis of Economic and Socio-Demographic Factors Affecting the Con- sumption of Food away from Home. Purdue Univ. Agricultural Experiment Station Bulletin No. 480, 1986. Menkhaus, D., G. Whipple, S. Torok, and R. Field, Devel- oping a Marketing Strategy for Branded, Low Fat, Fresh Beef, Agribusiness, 4(1988):9l-l03. Moschini, G., and K. Meilke, Parameter Stability and the U.S. Demand for Beef, Western journal of Agricultural Economics, 9(l984):27l-282. Nyankori, J., and G. Miller, Some Evidence and Implications of Structural Change in Retail Demand for Meat, South- ern journal of Agricultural Economics, 14(1982):65- 70. Pao E., S. Mickle, and M. Burk, One-day and 3-day Nutrient Intakes by Individuals - Nationwide Food Consumption Survey Findings, Spring 1977, Journal of the Ameri- can Dietetic Association, 85(1985):3 13-24. Pindyck, R.S., and D.L. Rubinfeld, Econometric Models and Economic Forecasts, third edition, McGraw-Hill, 1991 . Prochaska, FJ., and R.A. Schrimper, Opportunity Cost 0f Time and Other Socioeconomic Effects on away from Home Food Consumption, American journal ofAgri- cultural Economics, 55(l973):595-603. Putnam, 1.1., Food Consumption, Prices, and Expenditures, 1967- 88, Statistical Bulletin Number 804, Economic Research Service, U.S. Department of Agriculture, Washington, D.C., 1990. Putnam, 1.1., and M.G. Van Dress, Changes Ahead for Eat- ing Out, National Food Review. 260984): 15-17. Redman, B., The Impact of Women’s Time Allocation on Expenditure for Meals away from Home and Prepared Foods, American journal ofAgricultural Economics, 62,2(1980):234-37. l4 SAS Institute, S UGI Supplemental Guide, Cary, North Caro- lina, 1983. Skaggs R., D. Menkhaus, S. Torok, and R. Field, Test Mar- keting a Branded, Low Fat, Fresh Beef, Agribusiness, 3(l987):257-71. Sexauer, B., The Effects of Demographic Shifts and Changes in the Income Distributions of Food away from Home Expenditure, American journal 0f Agricultural Eco- nomics, 6l(1979):1046- 57. Wall Street journal, Lean and Mean: Hardee’s Joins Low- Fat Fray, July l5, I991, p. Bl. Appendix Maximum Likelihood Estimates of the Logit Models for Meats 15 Table A.1. Maximum likelihood estimates of the logit model for beef from FAFH. - Changes in Variable Estimate Std. error probability“ intercept -2.263" 0.610 -0.274 Urban1 0.136 0.105 0.016 Urban2 -0.022 0.089 -0.003 Region1 -0.212‘ 0.106 -0.025 Region2 -0.066 0.094 -0.007 Region4 -0.188* 0.107 -0.022 Race2 -0.447* 0.152 -0.054 Race3 0.197 0.375 0.024 Race4 0.196 0.257 0.023 Hisp1 -0.057 0.224 -0.007 Sex1 0.346* 0.073 0.041 Employ1 0.275‘ 0.082 0.033 Fstampi -0.020 0.209 -0.002 Diet1 0.004 0.109 0.0005 Hsize -0.158* 0.029 -0.019 Logage -0.293* 0.091 -0.0008 Logincome 0.186’ 0.051 0.76-06 Weekend 0.217‘ 0.093 0.026 Quarter1 -0.116 0.087 -0.014 Quarter3 -0.105 0.112 -0.013 Quarter4 -0.179* 0.107 -0.021 R statistic 0.1350 McFadden R2 0.0260 Likelihood ratio tests Urbanization 2.77 Region 5.47 Race 10.47’ Season 3.58 Number of iterations 5 Ratio" 0.1 487 ‘Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.1209, while the value of z is -1.8087. ° Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGI supple- mental guide, 1983 edition of SAS for further details. l6 Table A.2. Maximum likelihood estimates of the logit model for pork from FAFH. Changes in Variable Estimate Std. error probabilitya Intercept -2.892* 0.726 -0.245 Urban1 -0.285‘ 0.126 -0.024 Urban2 -0.182* 0.102 -0.015 Region1 -0.114 0.123 -0.009 Region2 -0.046 0.111 -0.004 Region4 -0.155 0.128 -0.013 Race2 0.144 0.157 0.012 Race3 -0.449 0.602 -0.038 Race4 -0.457 0.402 -0.038 Hisp1 -0.269 0.309 -0.022 Sex1 0.175* 0.086 0.014 Employ1 0.503* 0.101 0.042 Fstampi -0.079 0.259 -0.006 Diet1 -0.275* 0.136 -0.023 Hsize -0.226* 0.036 -0.019 Logage -0.156 0.109 -0.0003 Logincome 0.179’ 0.061 0.51 -06 Weekend 0.164 0.110 0.014 Quarter1 -0.119 0.104 -0.010 Quarter3 -O.129 0.134 -0.010 Quarter4 0.081 0.119 0.006 R statistic 0.1450 McFadden R2 0.0310 Likelihood ratio tests Urbanization 5.68 Region 1 .78 Race 2.98 Season 3.38 Number of iterations 6 Ratio” 0.1021 ‘Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.0848, while the value of z is -2.2696. l’ Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGl supple- mental guide, 1983 edition of SAS for further details. I Table A.3. Maximum likelihood estimates of the logit model for lamb, veal, and game from FAFH. Changes in Variable Estimate Std. error probabilitya lntercept -17.710’ 3.736 -0.039 Urban1 0.232 0.650 0.0005 Urban2 0.530 0.540 0.001 Regioni 0.598 0.490 0.001 Region2 0.036 0.564 0.0001 Region4 0.027 0.589 0.0001 Race2 0.100 0.805 0.0002 Race3 -5.809 5.901 -0.013 Race4 1.999’ 1.092 0.004 Hisp1 -7.1352 2.770 -0.016 Sex1 0.393 0.381 0.0008 Employ1 0.527 0.475 0.001 Fstamp1 1.952’ 0.879 0.004 Diet1 0.237’ 0.475 0.0005 Hsize -0.263 0.167 -0.0006 Logage 1 .084’ 0.576 0.00005 Logincome 0.735’ 0.293 0.55-07 Weekend 0.557 0.427 0.001 Quarter1 0.376 0.441 0.0008 Quarter3 0.227 0.598 0.0005 Quarter4 0.188 0.552 0.0004 R statistic 0.1880 McFadden R2 0.0820 Likelihood ratio tests Urbanization 1.11 Region 1.98 Race 2.39 Season 0.73 Number of iterations 8 Ratio” 0.0048 ‘ Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.0022, while the value of z is -6.1017. ° Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGl supple- mental guide, 1983 edition of SAS for further details. 17 Table A.4. Maximum likelihood estimates of the logit model for poultry from FAFH. Changes in Variable Estimate Std. error probability“ Intercept -1.452’ 0.650 -0.149 Urban1 0.097 0.111 0.010 Urban2 -0.048 0.096 -0.005 Regiont -0.375’ 0.117 -0.038 Region2 -0.096 0.101 -0.009 Region4 -0.177 0.114 -0.018 Race2 0.310’ 0.134 0.032 Race3 0.139 0.416 0.014 Race4 -0.119 0.308 -0.012 Hisp1 -0.019 0.251 -0.002 Sex1 -0.061 0.078 -0.006 Employ1 0.455’ 0.089 0.046 Fstamp1 -0.299 0.237 -0.030 Diet1 0.115 0.115 0.012 Hsize -0.250’ 0.033 -0.025 Logage -0.566* 0.097 -0.001 Logincome 0.218’ 0.056 0.76-06 Weekend 0.227’ 0.098 0.023 Quartert -0.162’ 0.094 -0.016 Quarter3 -0.254’ 0.124 -0.026 Quarter4 0.056 0.108 0.006 R statistic 0.1750 McFadden R2 0.0390 Likelihood ratio tests Urbanization 2.03 Region 10.89’ Race 5.40 Season 7.79 Number of iterations 5 Ratio“ 0.1281 ‘Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.1032, while the value of z is -2.0224. l’ Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGl supple- mental guide, 1983 edition of SAS for further details. Table A.5. Maximum likelihood estimates of the logit model for fish and shellfish from FAFH. ~ Table A.6. Maximum likelihood estimates of the logit model for ‘ beef from FAH. Changes in Changes in Variable Estimate Std. error probabilitya Variable Estimate Std. error probability“ Intercept -7.186’ 0.826 -0.520 Intercept -1.734’ 0.425 -0.431 Urban1 0.271 ’ 0.136 0.019 Urban1 -0.201’ 0.075 -0.050 Urban2 0.169 0.117 0.012 Urban2 -0.168* 0.062 -0.041 Region1 0.033 0.123 0.002 Region1 -0.104 0.075 -0.025 Region2 -0.141 0.121 -0.010 Region2 -0.070 0.068 -0.017 Region4 -0.337’ 0.139 -0.024 Region4 -0.061 0.076 -0.015 Race2 -0.154 0.185 -0.011 Race2 -0.271* 0.096 -0.067 Race3 0.726’ 0.419 0.052 Race3 -O.651’ 0.302 -0. 161 Race4 -0.046 0.396 -0.003 Race4 -0.239 0.183 -0.059 Hisp1 -0.054 0.312 -0.004 Hisp1 0.425’ 0.157 0.105 Sex1 0.081 0.091 0.006 Sex1 0.179’ 0.052 0.044 Employ1 0.403’ 0.107 0.029 Employ1 -0.183* 0.057 -0.045 Fstamp1 -0.326 0.341 -0.023 Fstamp1 -0.236’ 0.128 -0.058 Diet1 0.216’ 0.123 0.015 Diet1 -0.140’ 0.076 -0.034 Hsize -0.192* 0.039 -0.014 Hsize 0.162’ 0.020 0.040 Logage 0.374’ 0.123 0.0006 Logage 0.292’ 0.065 0.001 Logincome 0.356’ 0.068 0.8706 Logincome 0.025 0.034 0.21 -06 Weekend 0.327’ 0.113 0.023 Weekend 0.079 0.070 0.019 Quarter1 0.122 0.106 0.009 Quarter1 -0.118’ 0.062 -0.029 Quarter3 0.008 0.140 0.0006 Quarter3 0.291 ’ 0.079 0.072 Quarter4 -0.233* 0.141 -0.016 Quarter4 -0.090 0.075 -0.022 R statistic 0.1740 R statistic 0.1220 McFadden R2 0.0408 McFadden R2 0.0195 Likelihood ratio tests Likelihood ratio tests Urbanization 4.14 Urbanization 9.54’ Region 8.20’ Region 2.21 Race 3.44 Race 13.39’ Season 6.28 Season 26.68’ Number of iterations 6 Number of iterations 5 Ratio” 0.0894 Ratio” 0.4603 ‘Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.0723, while the value of z is -2.4623. "Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGI supple- mental guide, 1983 edition of SAS for further details. 18 *Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.2483, while the value of z is -0.1628. l’ Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGI supple- mental guide, 1983 edition of SAS for further details. Table A.7. Maximum likelihood estimates of the logit model for pork from FAH. Changes in Variable Estimate Std. error probabilitya Intercept -1.991’ 0.439 -0.463 Urban1 -0.251’ 0.078 -0.058 Urban2 -0.129’ 0.064 -0.030 Region1 -0.240" 0.077 -0.055 Region2 -0.053 0.070 -0.012 Region4 -0.341’ 0.081 -0.079 Race2 0.612’ 0.096 0.142 Race3 0.491’ 0.285 0.144 Race4 0.129 0.189 0.030 Hisp1 -0.112 0.162 -0.026 Sex1 0.283’ 0.054 0.065 Employ1 -0.118’ 0.059 -0.027 Fstamp1 -0.037 0.129 -0.008 Diet1 -0.221’ 0.079 -0.051 Hsize 0.120’ 0.020 0.028 Logage 0.473’ 0.068 0.002 Logincome -0.053 0.035 -0.42-06 Weekend 0.226’ 0.072 0.052 Quartert 0.028 0.064 0.006 Quarter3 0.017 0.081 0.004 Quarter4 -0.067 0.078 -0.015 R statistic 0.1440 McFadden R2 0.0255 Likelihood ratio tests Urbanization 10.48’ Region 23.11’ Race 42.47’ Season 1.42 Number of iterations 5 Ratio“ 0.3727 ’Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.2327, while the value of z is -0.5374. b Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGI supple- mental guide, 1983 edition of SAS for further details. l9 Table A.8. Maximum likelihood estimates of the logit model for lamb, veal, and game from FAH. Changes in Variable Estimate Std. error probabilitya Intercept -4.719’ 1.415 -0.101 Urban1 -0.307 0.234 -0.006 Urban2 -0.588’ 0.201 -0.012 Region1 0.883’ 0.233 0.018 Region2 0.222 0.239 0.004 Region4 0.287 0.267 0.006 Race2 0.280 0.305 0.006 Race3 0.730 0.736 0.015 Race4 -0.836 0.759 -0.017 Hisp1 0.411 0.457 0.009 Sex1 0.020 0.170 0.0004 Employ1 -0.097 0.185 -0.002 Fstamp1 0.609’ 0.344 0.013 Diet1 0.035 0.238 0.0007 Hsize -0.056 0.065 -0.001 Logage -0.064 0.211 -0.00003 Logincome 0.139 0.118 0.1-06 Weekend -0.240 0.246 -0.005 Quarterl 0.122 0.199 0.003 Quarter3 -0.326 0.293 -0.007 Quarter4 0.197 0.231 0.004 R statistic 0.1650 McFadden R2 0.0219 Likelihood ratio tests Urbanization 8.41 ’ Region 15.13’ Race 3.21 Season 3.26 Number of iterations 5 Ratio” 0.0241 ’Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.0213, while the value of z is -3.8026. °Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGl supple- mental guide, 1983 edition of SAS for further details. Table A.9. Maximum likelihood estimates of the logit model for poultry from FAH. Table A.10. Maximum likelihood estimates of the logit model for fish and shellfish from FAH. Changes in Changes in Variable Estimate Std. error probability“ Variable Estimate Std. error probability“ Intercept -3.143’ 0.451 -0.723 Intercept -3.777’ 0.543 -0.595 Urban1 0.156’ 0.079 0.036 Urban1 0.264’ 0.096 0.041 Urban2 0.172’ 0.066 0.039 Urban2 0.347’ 0.081 0.054 Region1 -0.109 0.078 -0.025 Region1 0.244’ 0.091 0.038 Region2 -0.205’ 0.072 -0.047 Region2 0.011 0.088 0.002 Region4 -0.201’ 0.080 -0.046 Region4 0.205’ 0.094 0.032 Race2 0.987’ 0.097 0.227 Race2 0.339’ 0.114 0.053 Race3 0.816’ 0.284 0.187 Race3 1.852’ 0.289 0.292 Race4 0.446’ 0.184 0.102 Race4 -0.940’ 0.298 -0.148 Hisp1 0.415’ 0.156 0.095 Hisp1 0.335’ 0.191 0.052 Sex1 0.005 0.055 0.001 Sex1 -0.042 0.065 -0.007 Employ1 -0.181’ 0.059 -0.041 Employ1 -0.077 0.071 -0.012 Fstamp1 -0.133 0.133 -0.030 Fstamp1 0.165 0.156 0.026 Diet1 ' 0.341 ’ 0.077 0.078 Diet1 0.459’ 0.085 0.072 Hsize 0.099’ 0.020 0.022 Hsize -0.018 0.025 -0.003 Logage 0.469’ 0.069 0.002 Logage 0.328’ 0.083 0.001 Logincome 0.040 0.036 0.31 -06 Logincome 0.097’ 0.044 -0.51-06 Weekend -0.017 0.074 -0.004 Weekend -0.135 0.089 -0.021 Quartert 0.129’ 0.065 0.029 Quarteri -0.106 0.077 -0.016 Quarter3 0.141’ 0.082 0.032 Quarter3 -0.206* 0.099 -0.032 Quarter4 0.075 0.079 0.017 Quarter4 -0.213’ 0.095 -0.033 R statistic 0.1750 R statistic 00.1490 McFadden R2 0.0355 McFadden R2 0.0284 Likelihood ratio tests Likelihood ratio tests Urbanization 7.33’ Urbanization 18.85’ Region 10.30’ Region 10.85’ Race 113.54’ Race 60.45’ Season 5.10 Season 7.50 Number of iterations 5 Number of iterations 6 Ratio” 0.3641 Ratio" 0.2039 ‘Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.2302, while the value of z is -0.5787. l’ Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGl supple- mental guide, 1983 edition of SAS for further details. ‘Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.1576, while the value of z is -1.4108. b Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGI supple- mental guide, 1983 edition of SAS for further details. ii \ Table A.11. Maximum likelihood estimates of the logit model for beef from all foods. Changes in Variable Estimate Std. error probability" Intercept -1 .478’ 0.422 -0.364 Urban1 -0.134’ 0.075 -0.033 Urban2 -0.146’ 0.063 -0.036 Region1 -0.157’ 0.075 -0.038 Region2 -0.097 0.069 -0.024 Region4 -0. 1 O3 0.077 -0.025 Race2 -0.339’ 0.095 -0.083 Race3 -0.367 0.284 -0.091 Race4 -0.047 0.186 -0.011 Hisp1 0.464’ 0.162 0.114 Sex1 0.269’ 0.053 0.066 Employ1 -0.046 0.057 -0.011 Fstamp1 -0.238’ 0.126 -0.058 Diet1 -0.160’ 0.075 -0.039 Hsize 0.100’ 0.020 0.024 Logage 0.217’ 0.065 0.001 Logincome 0.075’ 0.034 0.62-06 Weekend 0.137’ 0.071 0.034 Quarter1 -0.118’ 0.062 -0.029 Quarter3 0.234’ 0.080 0.057 Quarter4 -0.148 ’0.075 -0.036 R statistic 0.1120 McFadden R2 0.0171 Likelihood ratio tests Urbanization 5.84 Region 4.95 Race 13.89‘ Season 22.23’ Number of iterations 5 Ratio“ 0.5560 ’Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.2467, while the value of z is 0.2298. “Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGl supple- mental guide, 1983 edition of SAS for further details. 21 Table A.12. Maximum likelihood estimates of the loglt model for pork from all foods. Changes in Variable Estimate Std. error probability‘ Intercept -1.462’ 0.425 -0.361 Urban1 -0.242’ 0.076 -0.059 Urban2 -0.137’ 0.063 -0.033 Region1 -0.211’ 0.075 -0.052 Region2 -0.060 0.068 -0.014 Region4 -0.344’ 0.077 -0.084 Race2 0.577’ 0.095 0.142 Race3 0.286 0.283 0.070 Race4 0.071 0.184 0.017 Hisp1 -0.152 0.157 -0.037 Sex1 0.295’ 0.053 0.072 Employ1 0.029 0.057 0.007 Fstamp1 -0.049 0.127 -0.012 Diet1 -0.311’ 0.077 -0.076 Hsize 0.055’ 0.020 0.013 Logage 0.379’ 0.066 0.002 Logincome -0.028 0.034 -0.23-06 Weekend 0.227’ 0.070 0.056 Quartert -0.006 0.063 -0.001 Quarter3 -0.031 0.079 -0.008 Quarter4 -0.012 0.076 -0.003 R statistic 0.1260 McFadden R2 0.0206 Likelihood ratio tests Urbanization 10.62’ Region 23.20’ Race 37.1 1 ’ Season 0.16 Number of iterations 5 Ratio” 0.4426 ’Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.2465, while the value of z is -0.2360. l’ Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGl supple- mental guide, 1983 edition of SAS for further details. Table A.13. Maximum likelihood estimates of the logit model for lamb, veal, and game from all foods. Table A.14. Maximum likelihood estimates of the logit model for poultry from all foods. Changes in Changes in Variable Estimate Std. error probability“ Variable Estimate Std. error probability?‘ Intercept -6.111* 1.343 -0.157 intercept -2.202* 0.429 -0.546 Urban1 -0.259 0.220 -0.007 Urban1 0.166’ 0.076 0.041 Urban2 -0.436* 0.185 -0.011 Urban2 0.125* 0.063 0.031 Region1 0.839’ 0.211 0.021 - Region1 -0.210* 0.075 -0.052 Region2 0.181 0.220 0.004 Region2 -0.174* 0.069 -0.043 Region4 0.250 0.244 0.006 Region4 -0.237* 0.077 -0.059 Race2 0.247 0.285 0.006 Race2 1.036* 0.099 0.257 Race3 0.614 0.734 0.016 Race3 0.818’ 0.290 0.203 Race4 -0.407 0.636 -0.010 Race4 0.306* 0.182 0.076 Hisp1 0.164 0.459 0.004 Hisp1 0.403‘ 0.154 0.100 Sex1 0.088 0.155 0.002 Sex1 -0.029 0.053 -0.007 Employ1 -0.013 0.172 -0.0003 Employ1 -0.010 0.057 -0.003 Fstamp1 0.815* 0.323 0.021 Fstamp1 ~0.093 0.129 -0.023 Diet1 ' 0.075 0.214 0.002 Diet1 0333* 0.076 0.082 Hsize -0.081 0.061 -0.002 Hsize -0.011 0.020 -0.003 Logage 0.091 0.197 0.00005 Logage 0.214’ 0.065 0.001 Logincome 0.228* 0.112 0.19-06 Logincome 0.113’ 0.035 0.95-06 Weekend -0.069 0.212 -0.002 Weekend 0.057 0.071 0.014 Quarter1 0.170 0.181 0.004 Quarter1 0.034 0.063 0.008 Quarter3 -0.232 0.263 -0.006 Quarter3 0.023 0.079 0.006 Quarter4 0.194 0.214 0.005 Quarter4 0.042 0.076 0.010 R statistic 0.1270 R statistic 0.1480 McFadden R2 0.0206 McFadden R2 0.0265 Likelihood ratio tests Likelihood ratio tests Urbanization 5.45 Urbanization 5.80 Region 17.10’ Region 13.47* Race 1.80 Race 120.21 * Season 3.20 Season 0.45 Number of iterations 5 Number of iterations 5 Ratio” 0.0288 Ratio“ 0.4582 ‘Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.0257, while the value of z is -3.6057. b Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGI supple- mental guide, 1983 edition of SAS for further details. 22 *Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value of the logistic density function. At the sample means, the value of this density function (f(z)) is 0.2482, while the value of z is -0.1671. b Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGI supple- mental guide, 1983 edition of SAS for further details. Q i Table A.15. Maximum likelihood estimates of the logit model for fish and shellfish from all foods. Changes in Variable Estimate Std. error probability‘ Intercept -4.1 23’ 0.497 -0.806 Urban1 0.299’ 0.086 0.058 Urban2 0.313’ 0.073 0.061 Region1 0.222’ 0.082 0.043 Region2 , -0.039 0.078 -0.008 Region4 0.062 0.086 0.012 Race2 0.185’ 0.106 0.036 Race3 1 .579’ 0.292 0.308 Race4 -0.794’ 0.258 -0.155 Hisp1 0.288 0.178 0.056 Sex1 0.006 0.059 0.001 Employ1 0.074 0.065 0.014 Fstamp1 0.121 0.149 0.023 Diet1 0.441 ’ 0.079 0.086 Hsize -0.074’ 0.023 -0.014 Logage 0.362’ 0.075 0.002 Logincome 0.167’ 0.040 0.000001 Weekend 0.007 0.078 0.0003 Quarter1 -0.016 0.069 -0.003 QuarterS -0.173’ 0.090 -0.034 Quarter4 -0.231’ 0.086 -0.045 R statistic 0.1550 McFadden R2 0.0294 Likelihood ratio tests Urbanization 20.62‘ Region 1 0.46’ Race 43.95’ Season 9.99’ Number of iterations 5 Ratio" 0.2742 ‘Statistical significance at the 0.05 level. a Equal to the product of the parameter estimates times the value oi the logistic density function. At the sample means, the value of this density function (f(z)) is 0.1954, while the value of z is -1.0129. ° Ratio of nonzero observations to the total number of observations (6276). Note: The R statistic is similar to the multiple correlation coefficient in the normal setting, after a correction is made to penalize for the number of parameters estimated. See page 183 of the SUGI supple- mental guide, 1983 edition of SAS for further details. 23 %‘<'_.. [Blank Page m Orignai Bulletin] V w. [Blank Page in Original Bulletin] ' Texas Agricultural Experiment Station The Texas A&M University System Mention of a trademark or a proprietary product does not constitute a guarantee or a warranty of the product by the Texas Agricultural Experiment Station and does not imply its approval to the exclusion of other products that also may be suitable. ‘ All programs and information of the Texas Agricultural Experiment Station are available to everyone without regara to race, color, religion, sex, age, handicap, or national origin. Copies printed: 1,500