Research Methods: Assessment of Case Studies - Essay Example

RESEARCH METHODS: ASSESSMENT OF CASE STUDIES May 2006 RESEARCH METHODS: ASSESSMENT OF CASE STUDIES Introduction Modern theory and applied science demonstrate increasing sophistication of both research methods and demands to the research process itself. According to different sources, almost 90% of modern surveys in social, natural and applied sciences utilize different kinds of research methods and further computerized procession of data (Bishop, 2005). Also modern researches in different scientific domains are practically impossible without a comprehensive qualitative analysis in many aspects the methods employed are either used wrongly or have little practical effectiveness. As a result, many researchers reject empirical quantitative methods and make hypothesis based on their own vision of the problem that contradicts major principles of science, i.e.: verification of retrieved data. Another important problem is that a number of methods used in modern researches are not appropriate to each certain case. In this case the very idea of empirical survey is deteriorated and the results leave much to be desired. In both instances the retrieved results are far from reality and need substantial clarification. At the same time, properly selected and carefully employed research methods can be dramatically important in each type of either a survey or analysis of existing quantitative data. Correctly (from methodological perspective) received and interpreted data would provide a researcher with a variety of information and findings concerning different aspects of problem researched. The abovementioned makes the question of methods effectiveness and appropriateness to each certain case extremely important. This project will examine appropriateness and effectiveness of research methods suggested in each of three suggested case studies below. Case studies brief description The case studies provided for the analysis are dedicated to the same research problem, i.e.: examination of neighborhood effect on people's health behavior, particularly walking and smoking. All three case studies hypothesized that there is a significant influence of neighborhood on people's health behavior. However, each of them focused on identification of specific community factors that may influence this or that behavior. Each of these studies is based on representative samplings and includes the data from all-national (regional) surveys in the US or Canada. Alternative methods used were telephone surveys and observation of communities. Statistical methods in all three cases included different types of regressions that allows define the impact of one variable/variables (independent variable) on another one/ones (dependent variable). The project of C. Ross (2000) is dedicated to the general problem of health behavior. The author hypothesizes that neighborhood could affect health behaviors. Ross (2000) assumed that people from poor communities and with humble backgrounds (e.g.: poor education, low income, etc.) are more likely to smoke and reversely less likely to take exercises and walk. This assumption is based on two social-psychological models of influence in "community-person" system, i.e.: contagion theory and structural perspective. The first theory states that people are influenced by others surrounding them. The second approach believes that surrounding presents their neighbors with both constrains (fears, prejudices, etc.) and opportunities and resources (courts, playgrounds, pools, etc). The representative sampling (multilevel-data sampling) included the residents of Illinois (USA) selected according to the criteria of poverty and education of both the people and their neighborhoods. Other socio-demographic criteria included into the sampling were race, gender, age, household income, individual poverty, marital status, etc. The data for the sampling was taken from 1995 survey of Community, Health and Crime. The interviews were telephone-based and included both closed and open-ended questions. The retrieved data was analyzed by means of ordinary Lagrange Multiplier test (was insignificant in all the cases), least squares method and linear regression. The least squares method is ideal when there is a necessity to define proximity of variables within a single sampling. The linear regression method is optimal for definition of causality linkages between the variables (McCullagh & Nelder, 1989). Upon the analysis it was found out that also neighborhood does affect human behavior, this effect is not as significant as it was supposed. For example, both the residents of poor and rich communities were found to walk enough due to density of population, culture and transportation. The neighborhood was not found a predicative factor of exercising as well as smoking. The project of Craig et al (2002) explores the effect of the environment on leisure-time physical activity. The paper was aimed to analyze how the physical design of neighborhoods (e.g.: accessible appliances, social support, traffic value, etc.) influences people's behavior by disentangling effects of income, university degree, indexes of poverty and urbanization, and walking to work. The sampling included twenty seven Canadian neighborhoods different in income, social class, economic status and urban design including different urban areas with information of walking to work. The sampling included the residents of three Canadian provinces (both Canadians and legal immigrants); characteristics of neighborhood were rated by observer according to 10-point Likert scale. Hierarchical linear modeling was used to measure the relationships between neighborhood characteristics and walking to work. It was found out that higher environmental score of a community correlates with people's walk to the work regardless of community's location in an urban or rural area. Controlling for university education, income, and poverty did not influence the found relationships. The study by Berrigan and Troiano (2002) examines the association between urban form and physical activity in the US adults, particularly the age of the houses in urban environment and people's walking behavior. The authors hypothesized that older houses are more likely to have sidewalks; have denser interconnected networks of streets; and often combine business and residential areas. This combination makes older dwellings more accommodated for walking. The survey included such control variables as age, gender, race, income, education level and measure of limits to physical activity (activity limitations), as well as existing transportation system in the community, home age and patterns of walking behavior, etc. The retrospective survey was based on the data from the US National Health and Nutrition Examination Survey taken on representative sampling including 17.030 residents. The exploratory analysis was based on the method of logistic regression used to estimate odds ratios and 95% confidence intervals as well as to control for the effects of gender, race/ethnicity, age, education level, household income, and activity limitations. Logistic regression is a kind of a standard regression used when the dependent variable is a dichotomy, e.g.: success - failure. It was found out that environmental variables effect walking frequency and suggest that home age may be a useful proxy for features of the urban environment that influence physical activity in the form of walking. It was found out that walking is more typical to the communities with older homes in urban communities (excluding southern states of the US) while there was no significant correlation between the age of the houses and walking in the rural neighborhoods. However, home age was not associated with any specific type of leisure-time activity. Discussion Prior to the analysis of each study's methods and their appropriateness and effectiveness it is important to provide the general analysis of all surveys considering that they are dedicated to the same problem and employ comparable methods of statistical analysis. Regardless somehow contradistinct findings all three surveys are very comprehensive and employ a set of methods that allow regarding them as methodologically correct. All the surveys are retrospective in nature and are based on samplings including either all-national or regional population. According to Kalton (1983) retrospective surveys have several limitations. First of all, they measures events that happened in the past and do not provide a current estimate. Another limitation is that a researcher has to work with the data that already exist without a chance to add new variables into the array. This limits applicability of retrospective samplings to each certain case assessing different angles of the problem. As a result, utilization of retrospective surveys brings both strong positive effect (a great number of interviewees and possibility to generalize the data) as well as an adverse effect (practical impossibility or at least difficulty to re-interview the sampling or its part and refine the scope of research interests). Therewith, hereinafter I will avoid criticizing or commenting the original sampling as it was not taken by the authors of each certain study; instead, I will draw limitations or advantages of each survey based on the original sampling. Eventually, in each study a comprehensive statistical analysis was performed. This included subtypes of regression analysis necessary to define the effect of one variable on the other one or ones. This method is both appropriate and effective, though limitation of each three case studies with pure regression analysis might have remained some problems in the research field underexamined. Findings as for each statistical method appropriateness and effectiveness will follow in the analysis of each certain case study. Discussion of methods used in C. Ross (2000) study Dislike the other surveys which are medical in nature, Ross (2000) focuses more on sociological angle of the problem. As a result, she examines not the issue of health but rather health behavior and/or health intentions of the people. That determines research approach of the author who stays on two socio-psychological premises explaining the influence of a community on its member, i.e.: contagion theory and structural model. Reasoning from these premises Ross (2000) hypothesizes that disadvantageous communities (with lower income, poor opportunities and low education) may have negative effect on residents' health patterns. Reversely, upper communities may have better effect on people's health activity. The hypothesis seems to be completely natural taking into consideration sociological nature of the project and used methodology. Moreover, the theory of causal attribution or stereotypes (or even the common sense) prompts that disadvantageous communities are likely provide negative effect on the people. Another hypothesis requiring testing was which model: contagion or structural play more significant role in people's health behavior in different types of communities. This hypothesis, however, remained unproven. Ross (2000) is again completely true that testing of the abovementioned hypothesis requires multilevel data arrays (Creswell, 2003). These arrays should include such variables as poverty, educational level as well as other variables. Unfortunately, dislike the rest case studies, this one is based on purely Illinois sampling and the author does not mention if the sampling is representative for this state (not saying for the nation as a whole), what is it socio-demographic structure and if any probability tests (e.g.: Kettel's t-test) were performed. Moreover, even the glance on the sampling structure demonstrates that it was not representative for Illinois as, for example, according to the survey 16% of respondents lived in Chicago, while city's contribution to state's demographic is around 24% (Illinois, 2006). This, however, limits study's validity and does not allow generalizing the findings on the general US sampling. Also upon the analysis of the study one might do this keeping in mind significant extent of error (as it is indeed hard to assume that preferences in Illinois would different from the patterns in the other states), it is important to consider that the other studies (Berrigan and Troiano, 2002) demonstrated different impacts of neighborhood on people in the South and rest of the US territories. Even hypothetically, people in California may be more likely to walk due to better climate that Alaskans. However, it is in fact the only survey of the rest two that used re-interviewing of people. This helped specify the range of questions and make the overall survey more reliable. Nevertheless, the methodology of the survey (random selection of respondents and telephone survey itself) may be regarded as appropriate and correct though several limitations of the study can be mentioned again. The interviewees were asked to measure health behaviors of the people based on their verbal self-reporting. However, asking people about the numbers of walks per week an interviewee might have encountered with a social desirability bias as the respondents usually answer the question in a "socially-approved" manner. In this case people might have described the image of an "ideal self" instead of a realistic picture. Nevertheless, usage objective characteristics employed in the analysis (poverty, education, and racial and ethnic composition) can minimize the potential negative effect of the social desirability bias. Thus the study despite its limitations does provide precious empirical evidence. The author is completely correct in selection of data procession methods. Ross (2000) chose Lagrange multiplier test that was insignificant in any cases, least squares method and linear regression. Least squares is a mathematical optimization technique that attempts to find a "best fit" to a set of data by attempting to minimize the sum of the squares of the differences (called residuals) between the fitted function and the data. Selection of this method is appropriate and effective as it successfully minimizes the amount of data and makes it easier for qualitative interpretation (Dennis, 1977). However, selection of this very method among the battery of alternative techniques in the present study is not enough reasoned. Linear regression is the process of fitting the best possible straight line through a series of points. It is conventionally used to reduce a set of points to a simple mathematical relationship easy for interpretation (i) or deduce the underlying trends from a set of measurements that are expected, on theoretical grounds to follow a straight line (ii). In the other words, linear regression allows to find the trends in an array of data and find the causalities important for its interpretation (McCullagh & Nelder, 1989). Selection of linear regression is appropriate in the present case as far as the study attempted to explore the impact of independent variables (poverty, education, overall neighborhood influence etc.) on the dependent variables (health lifestyle). Regardless the utilization of comprehensive statistical methods the demonstrated results are somehow axiomatic (which, nevertheless, does not reduce the value of the research) in terms that people in poor communities walk more (due to more proximate destination points and absence of transportation) and people in big cities walk more than in suburbs and small cities/towns as in the small places it is often more problematic to get to the destination point without using a car. Paradoxically, environment has a strong effect on men's smoking while this was not valid for the women. Unfortunately, Ross (2000) did not manage to test the hypothesis of whether the contagion or structural model is more suitable for interpretation of neighborhood impact on people's health behavior. This, however, follows from Ross's (2000) methodology. In order to test both concepts it was important to include specific questions into the sampling. For example, as far as contagion model states that people are literally "inspired" by neighborhood it was necessary to include the questions of such kind: "to what extent does the community influence your behavior" or "do you tend to behave in a socially approved manner", etc. Absence of these questions did not allow testing this very hypothesis while their testing was key task of a survey sociological in nature. Thus, also the survey is very comprehensive and used appropriate methods it somehow lacks effectiveness both in developing of survey methodology (on the level of questionnaires) and hypothesis testing. In order to define other causalities and linkages in the survey as well as other implicit factors t would have been more advantageous to use such research methods as correlation analysis and factor analysis. Utilization of these methods would have made the survey both more illustrative and free from some methodological carelessness. Nevertheless, the study of Ross (2000) may be regarded as comprehensive and reliable. Discussion of methods used in L. Craig et al (2002) study The study of Craig et al (2002) analyses the issue of community impact on health behavior from a more structural than sociological and group pressure perspective. The hypotheses of the survey are the following: (i) does the neighborhood design influence respondents' walking to work and (ii) do neighborhoods' level of poverty, income, education, and degree of urbanization influence the relationship between environmental factors and walking to work The study of Craig et al (2002) is based on Canadian sampling seizing three most populated country's provinces of Ontario, Quebec and Alberta. In each of the states the researchers selected three types of communities, i.e.: an urban center, a nearby suburban center and small urban center. The survey is very comprehensive in nature and combines both interviewing (the data taken from interviewing all-Canadian representative sampling) and observation. This allowed Craig et al (2002) comparing and verifying the validity of subjective data (taken from respondents) with the objective variables (defined by observers according to modified Likert scale). In fact, even without attempting to test their influence on residents the survey of Craig et al (2002) included the variables that Ross's (2000) survey needed Thus, in terms of survey methodology and design as well as sampling definition this survey may be regarded as virtually faultless. The data was processed by means of hierarchical linear modeling. Hierarchical linear modeling may be used effectively to incorporate data from multiple levels in order to determine the impact of individual and grouping factors on some "individual level outcome" (e.g.: school factors on individual performance of each students) (Bentler & Satorra, 2000). Selection of hierarchical linear regression is completely appropriate and effective enough to the given case as far as the study attempted to explore the impact of independent variables (neighborhoods' level of poverty, income, education, and degree of urbanization as well as ecological ranking of neighborhood) on a dependent variable (walking to the work). The survey demonstrated that walking to work was significantly related to the neighborhood environment score and was not moderated by degree of urbanization. It was found out that higher environmental score of a community correlates with people's walk to the work regardless of community's location in an urban or rural area. Controlling for university education, income, and poverty did not influence the found relationships. Also the study of Craig et al (2002), as mentioned before, is virtually flawless, it is important to mention a certain limitation of study. From the data retrieved it is impossible to define whether the people were eager to walk to the work due to ecological factors or out of some other reasons as, say, personal circumstances or subjective preferences. In this case it would be better either to interview people from the observed communities in order to identify their motives for walking to work or include the control group and test their intentions for walking. This would have provided the study more reliability and made it a reference point for other studies combining retrospective data analysis (data taken from surveys) and observations or re-interviewing of sampling. Alternatively, Craig et al (2002) might have factorized the data in order to define the groups of ecological factors that contribute to people's decision to walk instead of giving a ride. This would have given more illustration to the survey. Discussion of methods used in D. Berrigan and R.P. Troiano (2002) study The study by Berrigan and Troiano (2002) that examines the association between urban form and physical activity in the US adults as well follows the ecological approach to analyze the reasons of people's health behavior in certain environments. Berrigan and Troiano (2002) attempted to assess the age of the houses in urban environment and people's walking behavior. As well as the rest of surveys' Berrigan and Troiano (2002) was a retrospective one based on the data from the Third US NHNES taken on representative sampling including 17.030 residents. The data accessible from the survey included: measure of walking behavior, frequency and specificity of leisure-time activity, rural or urban location, demographic characteristics of each respondent (age, gender, race, income, education level) and measure of limits to physical activity (activity limitations) of each respondent. Unfortunately, the sampling used was not split by states and has only four "sub-samplings" representative for Northeast, Midwest, West and South. This limits study's findings, as the states within these four territories may not be homogeneous in respect to health behavior (Creswell, 2003). For example, people in southern California may have healthier lifestyles than those living in northern Washington or Oregon. The exploratory analysis was based on the method of logistic regression used to estimate odds ratios and 95% confidence intervals as well as to control for the effects of gender, race/ethnicity, age, education level, household income, and activity limitations. Selection of this model is appropriate as logistic regression is conventionally applied when the dependent variable is a dichotomy and the independents are of any possible type (Hosmer and Lemeshow, 2000). Taken into consideration that dependant variable was represented by a dichotomy "urban - rural neighborhood" there are no objections for choosing this very statistical method. Statistical data procession allowed finding that men are more likely to walk then women (i); non-Hispanic white sample are more eager to walk then the rest respondents (ii); high school education and less then high school education were associated with rare walking (iii). The age of the houses built in the urban was found to be a factor that contributes to residents' walking. However, this was not a predictor when the houses were built in rural areas. This result was not affected by inclusion of control variables, i.e.: gender, income, ethnicity, education, etc. At the same time, home age was not found a predictor of other leisure-time activities such as running/jogging, swimming, dancing, etc. This however, may be a result of study's limitation. Also Berrigan and Troiano (2002) state that the results of the survey are consistent with the transportation survey the received data might have been interfered with a certain survey artifact. The intention of older houses' residents to walk more might be explained particularly with worse transportation in the district. Moreover this assumption needs clarification that the age of the house was not associated with any specific type of leisure-time activity. In order to avoid this potential bias it might have been important to provide an observation of selected neighborhoods in each certain locale or, alternatively, choose a control group. In this case Berrigan and Troiano (2002) would have defined for sure that it is the house's age that influences people's walking instead of, say, poor transportation in the district. Another limitation of the study is that it does not explain the association between the age of houses and each of four selected reasons. The fact that this factor does not predict walking in southern states requires more profound analysis and interpretation. Nevertheless, the overall study can be regarded as methodologically correct. The methods of survey design and data procession are appropriate to the case and effective in provision of reliable results. Bibliography: 1. Berrigan D. and Troiano R.P., 2002. The Association between Urban Form and Physical Activity in U.S. Adults American Journal of Preventive Medicine, 23 (2S): 74-79 2. Craig C.L., Brownson C.R., Cragg, S.E. and Dunn A.L., 2002. Exploring the Effect of the Environment on Physical Activity: A Study Examining Walking to Work Am J Prev Med 23(2S): 36-43 3. Ross C.E., 2000. Walking, exercising, and smoking: does neighborhood matter Social Science & Medicine 51: 265-274 4. Bishop G.F. 2005. Illusion of Public Opinion: Fact and Artifact in America's Public Opinion Polls. Oxford: Rowman and Littlefield Publishers 5. Creswell J.W., 2003. Research Design: Qualitative, Quantitative, and Mixed Methods Approaches. Sage Publication, thousand Oaks, CA. 6. Kalton G., 1983. Introduction to Survey Sampling. Sage Publications 7. Illinois 2006 [online] http://en.wikipedia.org/wiki/Illinois accessed May 10 2006 8. Hosmer D.W. Jr., Lemeshow S. 2000. Applied Logistic Regression. New York: Willey 9. McCullagh P., and Nelder J. A., 1989. Generalized Linear Models. London: Chapman and Hall. 10. Bentler P.M. and Satorra A. 2000. Hierarchical regression without phantom factors. Structural Equation Modeling: A Multidisciplinary Journal, 7(2), 287-291. 11. Dennis, J.E., Jr. 1977, Nonlinear least-squares. State of the Art in Numerical Analysis D. Jacobs (Ed.), Academic Press, 269-312. Read More