Sorin Matei
Annenberg School for Communication
University of Southern California
1443 S. Barry Ave. # 206
Los Angeles, CA 90025
Phone: 310 312 2973; Fax: 603 737 6859
email: matei@usc.edu
personal webpage: http://matei.org
project webpages: http://www.metamorph.org * http://www.metamorph.org/maps.
This paper describes the process of GIS modeling used for explaining maps of social perception of space generated by selected groups of Los Angeles county residents. It also tests the statistical significance of the association between these perceptions and several factors that shape them, especially housing desirability, racial relations and media connectedness.
The perceptual maps used in this study are based on respondents' feeling of "comfort" about the Los Angeles urban area in which they live. These feelings are captured by asking people to indicate, using colors, in what areas of a black and white map of the southern part of Los Angeles county they feel comfortable. The first underlying affective dimension I tried to get at was people's mental image of a "good area," a place they consider secure and desirable. The second element was the fear element in people's relationship with their social environment. Thus, I assumed that people's mental maps have a push and a pull component, both of affective nature.
"Comfort" was believed to be related to "pull" elements, to factors that make an area attractive, especially housing desirability. These areas are characterized by perceived high civic, entertainment and economic value. "Discomfort" is related to factors that keep people away from a place, which in an atomized urban environment such as Los Angeles always involves fear. This fear could be elicited by a variety of factors. This paper will use the explanatory paradigm used in communication studies which suggests that media "framing" of social reality has an important role in the process (Scheufele, 1999).
Media images are included in this study because they are central in articulating people's understanding of where they stand relative to social reality. One way or another, people's perception of an area is related to a communication process of some sort. The mental maps which congeal these impressions are thus to be seen as the product of communication, mass mediated or interpersonal.
Obviously, communication processes have their own way of encoding reality, which often involves interpreting it. This means that the reality in our heads, resulted from communication, is rarely an accurate image of objective reality. This is of crucial importance for the present study because I suspect that whereas some areas could be justly perceived as unsafe or highly desirable to live in, the image of many other areas is distorted through a process some media scholars designate by the concept of "framing" (Scheufele, 1999). Although this is a complex process, it is largely recognized that framing is part of the meaning-giving capacity of the human mind, which sets all "facts" on a canvass of relationships with prior knowledge, values and beliefs about the world (Cassirer, 1944; Langer, 1948). This frame of mind is shared socially by certain groups, defined ethnically, politically or socially (Gitlin, 1980; Scheufele; 1999). These mindsets have a specific need for coherence; they have a "story" to tell, which sometimes is based not on the facts but on the idea the group has about the ideal state of the world, about what people or actions are acceptable or about what models and values should be used during social action. In this process of valuation mass media plays a crucial role. Media is one of the central mechanisms by which meaning and understanding is produced in society and has the capacity to influence our perceptions of social reality, including those with a spatial dimension.
Over the past 40 years there has been substantial work done, especially in geography and psychology, on mental and perceptual maps (Carter, 1979; Lynch, 1960; Kitchin, 1997). Often, however, these studies took a narrow psychological view trying to explain the images people use to orient themselves in space in terms of individual characteristics. The mental mapping mechanisms were construed in individual-psychological terms. For example, Carter's (1979) study of the criminal's image of the city used personal construct theory, proposing that perceptual maps are the result of a simple process of mental abstraction that organizes space along instrumental attributes, mainly related to spatial orientation (Kelly, 1955).
Although not rejecting the psychological approach in its entirety, this study proposes that geographic perception, like other types of social images, is also cultural and social in nature. Geographic images in people's heads are determined not just by instrumental-personal orientation goals but they also carry cultural meaning. Geographic perception of space is a personal and cultural construct at the same time (Lynch, 1960).
Thus, mental maps should be seen as cultural maps. One of the most important roles in "producing" the maps is played by cultural meaning acquired through media. A growing body of media research seems to indicate that mass media, especially commercial electronic media, such as broadcast television, have the tendency to create social images that are structurally biased in their coverage of reality (Gerbner, 1994). This bias was identified by the cultivation school as the "mean world syndrome" (Gerbner, 1971, 1994). In essence this concept proposes the view that mass media propagates images of a violent and out of control world which instills in people's minds the idea that each person or social/ethnic group is in a state of constant warfare with other persons or groups. This syndrome increases the likelihood of inter-ethnic stereotyping and conflict (Gerbner, 1994).
Although not directly concerned with space, both framing and cultivation theories offer valuable insights for understanding how social space is constructed in the presence of modern mass media.
Research questions
The operational goals of this study can be summarized in the following research questions:
Are Los Angeles mental maps of comfort an accurate description of reality?
What is the goodness of fit between perceptual maps of comfort and spatial distribution of socio-demographic variables that are hypothesized to model them: housing desirability, crime, ethnic distributions?
What role does media play in molding these maps?
The first two research questions refer to the degree to which spatial distributions of mental or socio-demographic variables are similar or not. I start from the assumption that although they should be related to a certain degree, there will be significant gaps between them. The expectation is that, for example, the goodness of fit between crime occurrence and discomfort (fear) to be weak and in consequence that many areas to be feared on other grounds than lack of safety. A good explanation would be stereotypical images about the populations living there, which are in turn influenced by media connectedness.
Thus, the research questions should be turned into more specific hypotheses.
A first one is that comfort is not associated with crime:
H1: Areas that are perceived to be more comfortable than others are not also safer..
Comfort, on the other hand, seems to have more to do with inter-ethnic relations, especially perception of threat coming from minority groups. These perceptions in Los Angeles have taken dramatic turns twice in the last decades, during the 1965 and 1992 race riots. Since the population involved in these riots was mainly the poorer segment of the African-American group, I will hypothesize that discomfort (fear) will be associated with presence of African-American population.
H2: The higher the African-American population density the less comfortable the area will be perceived.
This perceptual bias against African-American population, as mentioned above, is not only due to the fact that this group resides in areas that have known two violent racial conflicts and higher poverty rates and relatively higher crime. I also hypothesize that the perceptual distortion comes from media coverage of the areas, especially by mainstream commercial television. Thus, another hypothesis will be that:
H4: More intense use of television for social orientation will be associated with higher levels of fear.
Comfort maps also have attractive areas. A fifth, and last, hypothesis refers to the association between comfort maps and "positive" social indicators, such as housing desirability. I expect housing desirability to be relatively highly correlated with comfort.
H5: Areas perceived as comfortable are more likely to be have high real estate value.
If this hypothesis is confirmed, then it will follow that people are more accurate in their assessments of the "good" areas of the city than of what they consider to be the "bad" ones.
Method
Map Collection
The main layer used for this study, comfort maps, is the result of combining individual maps of comfort. They were provided by the Metamorphosis project, an ongoing project at Annenberg School for Communication, University of Southern California, which looks at communication flows and sense of community in Los Angeles. Two hundred and fifteen people were recruited from among 1500 hundred respondents to a telephone survey conducted between 1998 and 1999 in six different neighborhoods of Los Angeles.
The neighborhoods were selected on ethnic-geographic criteria and are situated within 10 miles of the Los Angeles civic center (see Figure 1 ). Four neighborhoods were dominated numerically and culturally by one ethnic group: whites on the Westside and in South Pasadena, African-American population in greater Crenshaw, and Mexican-Americans in East Los Angeles. Two of the neighborhoods are inhabited by a variety of groups, but are defined by a specific ethnicity: Koreans in Koreatown and Central Americans in Pico-Union. Only respondents from the dominant ethnicity were selected in each area. In addition to answering to questions about sense of attachment to local community and communication patterns, they were asked to color in, using crayons, a paper map of the southern part of Los Angeles county depicting in black and white the highway grid of the area. The coloring exercise took place after the phone interview, either during face to face meetings or at the respondents' residence, in the later case the maps being mailed in to them. In both situations the participants were instructed to color in black their neighborhoods, in green the areas they feel comfortable about, in yellow or orange the areas they feel somewhat comfortable about, in red the areas they feel uncomfortable about and in blue the areas they do not know. Areas left blank were were assumed to be unknown and were assigned the color blue.
After coloring, the maps were redrawn manually into ArcView 3.1 as vector maps, using the "edit theme" function. Each contiguous area on the map colored in a specific way was translated into a polygon associated with a "comfort" integer value. Red areas were assigned a value of - 1, blue areas a value of 0, yellow/orange areas a value of 1 and green areas a value of 2. Thus, sense of comfort was translated into a ratio scale.
The vector maps were in turn used to generate raster maps. The decision to use raster maps was based on the fact that they allow conflating individual maps into an average map using map algebra. Raster maps are divided into small square cells, equal in size, created by dividing the map into rows and columns. In our case, the maps were divided into 280 columns and 290 rows. Each map has the exact same grid structure, individual cells corresponding in position for all individual maps. Each cell inherits the value of the vector map polygon it was generated from. This value varies from map to map.
The average map is obtained by adding the values of each cell located at the same coordinates in all individual maps and then by dividing this value by the number of maps. Various individual maps can be generated by dividing the respondents in different groups (ethnicity, media use, etc.).
The resulting average maps are not distinct polygon maps, characterized by integer values, as it was the case for the individual maps, but continuous coverage maps, the value of each cell changing by a greater or smaller fraction ranging from - 1 to 2.
Three average maps were generated. The first map is the result of averaging all available individual maps into a synthetic map of comfort in Los Angeles. This is the average spatial perception 215 respondents, representing 3 races and six different residential areas, have about comfortable and uncomfortable areas in Los Angeles. This map, was used for measuring the goodness of fit between comfort and socio-demographic variable distribution (see Figure 2 ; shades of red and green represent negative and positive deviations from the mean - dark green and burgundy are usually 2 or more standard deviations away from the mean).
For measuring the bias in comfort due to media connectedness two other average maps were created. These maps were obtained by first splitting the respondent sample into two groups: weak television connectors and strong television connectors. Based on data recorded during the phone interview, the first group was constituted of those who declared that they do not use television as a primary means of information for accomplishing major social tasks, such as learning about their community, buying products or for entertainment. The second group was composed of those who declared that they use television for reaching at least one of these goals (Ball-Rokeach, 1985). The maps created by the members of these two groups were averaged separately so that one map represented the view of those who are strongly connected to the world through television (see Figure 3) and the other one of those who are not using television for achieving personal and social goals, thus being weak television connectors (see Figure 4; Similarly to the previous comfort map shades of red and green represent negative and positive deviations from the mean - dark green and burgundy are usually 2 or more standard deviations away from the mean).
Geo-spatial components of the study
The study area is the southern part of Los Angeles County, including the eastern part of San Fernando Valley. The area was selected based on the assumption that the space one has usable knowledge about is limited by the boundaries of the urban geography in which he or she lives. As most of our respondents live in the central area of the county, I assumed that their daily forays will be limited to the north by San Gabriel Mountains, to the West by the geographic limits of the city and to the south east by distance. As seen in Figure 5, the area stretches from San Fernando to the north west to Cerritos / Long Beach to the south east, and from the coastal area, to the south west to Monrovia / El Monte to the north east. This area is focused on the central metropolitan area of Los Angeles county. The area had in 1998 a total population of 7,218,612, out of which 3,473,130 were White, 3,366,800 were Hispanic, 952,191 were Asian and 862,183 were black. The ethnicities selected for our study represent 90% of the county population.
Socio-demographic variables
The other variables used in the project are crime occurrence, home sale price and number of units sold in 1998, black population in 1998, population diversity, income and population stability. These variables were available at three levels of geography: census tracts, zip code areas and municipalities.
Occurrence of crime was obtained at the municipality level of geography. The features of this thematic maps are the administrative districts of Los Angeles Police Department (LAPD) and Los Angeles Sheriff's Department (LASD), the two law enforcement organizations operating in Los Angeles county, plus a smaller number of municipal police districts.
Part of the crime statistics were extracted from the LA Sheriff's Department Year in Review (LASD, 1997), another part were downloaded from the Department of Justice's Bureau of Statistics (CJSC, 1998) and from the Los Angeles Police Department (LAPD) website (LAPD, 1998). Crime data is provided in the format of the FBI crime index, which is based on the number of felonies committed during a specific year (in our case 1997) in a geographic area including: homicide, rape, aggravated assault, robbery, burglary, theft, grand theft auto, arson. The data is normalized by population, so that the final index score used in regressions was crime per capita.
The data was provided in aggregated format for each of LAPD's 18 districts, for independent municipalities in the county, for some un-incorporated areas of the county and for cities served under contract by the Los Angeles Sheriff's Department. The shape files associated with crime data were downloaded from the California State University of Los Angeles http://csars.calstatela.edu ). Because the municipality map of Los Angeles and the LAPD divisions were stored in two different files, they were combined into a new theme by redrawing some of the contours manually.
Race and ethnic distributions were mapped at census tract level. Census tracts are areas between a few to tens of square miles, bordered by major streets used by the Census Bureau for conducting and reporting its decennial survey of US population. The layer was acquired from a commercial vendor (GDT). The attribute data associated with the shape file was retrieved from the US Census Bureau's web site. However, because the census data is already 10 years old, for black population was used projection data purchased from a commercial vendor (CACI).
Race variables were used in two versions. Besides percentage black population there was also used a diversity index, representing the racial "entropy" index for each census tract. This measure, based on Shannon's measure of diversity, indicates on a scale between 0 and 2.7 how mixed (racially and ethnically) an area is (Turner & Allan, 1989). It was used in this study to see if the heterogeneity of the population in a census tract has the same importance as presence of black population in eliciting feelings of fear.
Urban studies and criminology literature frequently mentions population instability and poverty as factors conducive to urban decay (Morenoff & Sampson, 1997; Sawicki, 1996; Rountree & Kenneth 1996; Myers, 1998; DeFrances and Smith, 1998). To make sure that the bias in the comfort maps is due to hypothesized ethnic stereotypes and not to other social-infrastructural factors, the census bureau measure of population stability - percentage residents living in the same house over a period of 5 years - and census tract median income was also introduced in the datasets and in the statistical models.
Housing desirability information was collected from Los Angeles Times (1998; Tamaki, 1999). Each year in January the newspaper publishes a metropolitan Los Angeles list containing number of homes sold and median home sale price in each zip code area. The increase or decrease of median prices between 1997 and 1998 indicates how desirable housing is in a specific zip code area (see Figure 6 ). The zip code shape file was also acquired from GDT.
The analysis was performed at municipality / LAPD division level of analysis. Since some of the variables were available only at census tract level, they were aggregates at municipality level by converting the shapefile into a grid file and then by using the "summarize" function in ArcView Spatial Analyst. A similar procedure was used for assigning comfort values to each municipality / LAPD division. The analysis for the two television comfort maps was done using zip code area level of analysis, where crime data was translated into a gird file and then summarized into zip code areas. These limitations entailed by these procedures are addressed in the conclusions.
Statistical Analysis
The statistical modeling process e The main research questions, as mentioned above, address the issue of the degree of association between feelings of comfort and real and imagined safety and housing desirability in Los Angeles county. This was done by using spatial statistical techniques. As it is well known, performing analysis on spatial units should take into account the fact that cases are not randomly selected, they are not extracting like ball from an urn. Rather, they are picked together with the clusters of neighbors surrounding them, in Cressie's (1993) plastic formulation, in bunches, like grapes. This requires that correlation studies should be done accounting for spatial bias. This is similar to the requirement of time series analysis to remove first the trend in order to account for the influence of an event on those that succeed it.Statistical analysis was performed using the Splus SpatialStats plug-in for ArcView GIS, which allows to use feature tables in ArcView as data frames for the Splus statisical package (Kaluzny et al., 1998). This module performs spatial regression by creating first a neighborhood matrix, which accounts for spatial covariation of cases. Two types of analyses were performed: autocorrelations and spatial regressions. Autocorrelations reveal if patterns in a specific thematic distribution display a spatial trend, or not. Otherwise said, it measures if the spatial patterns in maps are random or not. If the patterns are not random, then one would expect that the trends to be explainable by one or more other variables. Spatial regressions teases out exactly this relationship, by looking at spatial distributions that covary across themes.
Spatial multiple regression takes into account space when measuring covariance of variables. This is accomplished by weighting each case according to the number and identity of its spatial neighbors. According to the authors of SpatialStats, the program used for performing the analyses for this study "The spatial process is modeled by predicting the outcome for each region based on its dependence on nearby or neighboring regions. If two regions are neighbors then random processes measured at these regions might be spatially correlated." (Kaluzny et al., 1998, p. 111). A neighborhood matrix is thus built, which is then used in calculating the covariance model. Similarly to time series analysis, the covariance matrix could take three forms: simultaneous (SAR), conditional (CAR) and moving average (MA). The CAR matrix is usually recommended, although this requires that the relationships between the geographic units should be symmetrical (i.e. - relationships between units are bi-directional). The differences between covariance models could be significant and sometimes only trial and error procedures can tell which one is the best option (Kaluzny et al., 1998, p. 129).
Finally, the regression models were used to generate new variables based on the residuals resulted from fitting the values of the variables to the regression line. The residuals were plotted on the map, the resulting distribution indicating where a variable is over or underestimated, relative to the predicted values. For example, regressing housing desirability on comfort has produced a fitted value for housing price increases for each geographic unit, based on its association with comfort. The housing desirability score of each case was subtracted from the fitted value, obtaining a "residual" negative or positive score. In case the value is negative, the deviation of the comfort score indicates that for that geographic unit the level of affective comfort is lower than what housing desirability predicts. This could be due to measurement error or due to other factors, unexplained by the model, yet. This is an invaluable tool for the policy analyst, who gets at a glance the image of under and over-estimation of one variable based on another variable.
Analysis
The analysis was performed in two steps: exploratory and descriptive-inferential. The exploratory phase involved map generation and autcorrelation analysis for identifying the possible variables associated with comfort.
Autocorrelations
Before assessing correlations between maps one should estimate the degree to which variable distributions across space is non-random. This will tell if the variables have a coherent spatial structure. This can be measured by performing an autocorrelation test. Autocorrelation measures "whether, and in what way, adjacent or neighboring values tend to move together" (Bailey, 1994; Griffith, 1996). Also, autocorrelation is required as a first step prior to any multiple regression analysis to determine if a spatial model or a simple model should be applied.
Table 1 shows that all except for one of the target variables are indeed autocorrelated at the 0.05 level of significance. The statistic used here, Moran's I coefficient, is a measure ranging between - 1 and 1, indicating, when extreme, that the variable is highly correlated with space, in a positive or a negative way. That is, when the value is 1 similar values are clustered together; when the value is - 1, dissimilar values are clustered together. When the values are randomly distributed the value is 0 (Barber, 1988).
The only variable that is not autocorrelated here is crime. This confirms prima faciae the hypothesis that crime is a less spatially predictable factor than expected, which also decreases the probability of being well correlated with other variables, such as comfort.
On the other hand, the highest autocorrelated variables are comfort, black population and housing desirability. This insight will help in our regression model building showing which are the variables to be introduced in the model testing the media influence hypothesis.
Bivariate regressions
Bivariate spatial regressions between comfort and the other target variables tell to what degree there is a gross relationship between them, before controlling for other factors.
Table 2 reveals that comfort is correlated with housing desirability and percent black population. Crime, is not significantly associated with comfort. Also, none of the indicators of social alienation predict comfort, negatively or positively, as neither does income.
Multivariate models
The association between comfort and socio-demographic variables was tested using multiple spatial regression. Table 3 presents unstandardized betas for variables entered in models (red values are significant at 0.05 level). Three full models including all six independent variables selected in the theoretical section, were generated, one for each type of neighborhood matrix (SAR, CAR and MA). In addition, a simple linear model was generated, using al six independent variables, and two reduced models, including only the variables that had a significant impact on the dependent variable (comfort) in the full model (housing desirability, crime, percent black population), plus income.
Hypothesis 1 is supported by all three full models: crime is not associated with comfort either positively or negatively. Although unexpected, the fact that the linear and the reduced models indicate that crime is correlated with comfort does not entirely contradict the hypothesis. This seems to indicate that in fact comfortable areas are more likely to be present in areas with relatively higher levels of crime, controlling for housing desirability, percent black population and median income. The most plausible explanation, suggested by comparing the crime (see Figure 7) and comfort (see Figure 2) spatial distributions is that with the exception of the white respondents, most of the subjects live in relatively high crime areas, which they tended to perceive as being more comfortable.
Hypothesis 2 is confirmed by the reduced CAR 1 model and by the SAR matrix full model. Areas that are perceived as threatening have higher proportions of black population. More importantly, the Black population effect in the full SAR model seems to be independent of the unique contribution of crime per capita or of that of housing desirability to feelings of comfort. The relationships also hold for a reduced SAR model (not shown in Table 3), the betas being similar in size with those in the CAR model.
Hypothesis 3 is confirmed by data displayed in table 4.
Column one displays the betas for the model predicting spatial distribution of comfort for people who don't use television as a primary social tool. Column two displays the model for respondents who use television as a primary tool. The beta for black population in the second model is greater than in the first one by 25%. This indicates that the fear increases by 25% for those who use television for accomplishing social goals. The effect becomes even more obvious when re-examining the comfort maps for the two groups (see Figure 3 for those who use TV and Figure 4 for those who do not use TV). As can be easily noticed, the television group map has a larger red area in its center (inhabited primarily by African-Americans) and the deviation from the mean of those values is far greater. Also, the green values become smaller and the surfaces of areas of intense comfort decrease.
Hypothesis 4 is the one which is confirmed consistently by all models in table 3. Housing desirability goes with perceived urban comfort.
Besides the consistent relationship between housing desirability and comfort, one should also note the wide disparities between the linear regression model and the spatially regressed models, especially for crime per capita. Although crime per capita is one of the most random variables in the dataset, in the linear model is highly correlated with comfort, although none of the spatially regressed models captures this relationships. This demonstrates the value of this procedure, over the classical, linear one. It is also clearly noticeable that the spatial models vary quite widely between themselves, according to the covariance matrix used.
Residual Maps
The results of the bivariate or multivariate regressions provide not only insight in the degree of association between two variables, they can also provide valuable tools of social and policy diagnosis. One of such tool is the residual map. This type of map is based on the values that result from the difference between the real value of a variable for each geographic unit and the value predicted for it by other variable through regression analysis. If the value is negative, then the predicted score was higher than the actual value, if it is positive the relationship is reversed. In other words, plotting the residuals indicates which geographic feature attributes are over or underestimated relative to other characteristics.
Figure 7 displays areas which are over or underestimated in terms of housing price relative to comfort. In green areas price increases are under the level predicted by people's perception of comfort toward them. In other words, they are good areas where the price increases were not as high as one would expect based on their level of comfort (thus of subjective desirability). Red areas are overestimated. The price increase cannot be justified by their level of comfort. In white areas house price increase keeps pace with level of comfort. This map can be used by the policy or by the business analyst for spotting promising or problematic areas. For example, Palos Verde, Manhattan Beach and Altadena seem to be overvalued relative to their level of comfort. In contrast, a good number of communities in South Central Los Angeles and toward the eastern border of the county seem to be undervalued.
Obviously, this procedure does not provide an "objective" measurement. It only tells us, given that two variables are associated, to what degree one strays from the best fit line. It also tells us how perceptions are modeled by biases and what are the possible hotspots in this process of distortion. For example, the residual map indicating over or under real estate valuation relative to level of comfort could in fact indicate that other factors might influence the demand for housing in these areas.
Discussion and conclusions
This research project has shown how mental maps of a portion of Los Angeles county fit the socio-demographic reality to which they refer. It has tried to describe and explain how mental images of security and fear in Los Angeles are generated.
A first finding is that comfort maps are best explained by housing desirability. The authors of the maps were looking at the maps, first and foremost, to identify the areas that have the most "pull," areas rich in resources and desirable to live in. People's perception of these areas seems to be accurate. Their instinctive image of the city is verified by the direction in which the real estate market goes. Another preliminary conclusion is that although the positive pole of the comfort maps seems to be somewhat accurate, the negative pole is less accurate. Among the potential factors that could've explained lack of comfort: race and media connectedness emerged as a significant factor. Presence of minorities, respectively black population, seems to elicit feelings of fear and discomfort on a consistent basis. This fear seems not to be correlated with other objective factors, such as crime or poverty. The tentative conclusion is that stereotypical images about the Black population of the city shapes the comfort maps.
Also, those people who connect to the world through television for achieving major personal goals seem to be more fearful of black inhabited areas. Our preliminary conclusion is that mass media might contribute to create an environment of inter-ethnic fear.
These findings, although interesting, should be taken with a certain degree of caution. The dataset and the maps raise a series of methodological problems. The most important one is that the analysis was conducted at a very gross level of resolution (municipalities and police reporting districts). This has not only offered imprecise locations for crime occurrence but has also has forced us to use re-aggregated socio-demographic data (income, Black population) by using the summarize function in ArcView Spatial Analyst. This method is known to offer only a gross estimation of the real values. The comfort scale is also somewhat imprecise and there is a lot of work that should be done before ascertaining its reliability. Also, the location bias of the comfort maps is another issue that should be solved and which could greatly improve the quality of the analysis.
There are also several issues related to the quality of the data, especially that of the comfort maps. A main problem with mental maps is that their quality varies with each respondent's ability to read maps and knowledge of the area mapped. This covaries with education, travel experience and interest in the area. From analyzing the degree to which people have followed the instructions on coloring in the maps and from their ability to indicate consistent patterns of comfort/discomfort in Los Angeles area it was quite obvious that certain respondents could recognize only the names of the areas, which were indicated as comfortable or not, but not a specific geographical extent of those areas. However, the fact that average maps display clear spatial patterns indicates that most of the respondents had a pretty coherent mental image of the city.
The way in which crime data is provided can also rise some issues. Crime statistics are provided in the aggregate, for rather large geographical units, especially in the city of Los Angeles. This creates a difference in resolution between the crime map and the comfort map, which makes the predictive model less precise than desired. Also, representing crime occurrence as a global indicator for a whole polygon tends to ignore the fact that certain areas are more prone to crime then others, even within specific areas. In a best world scenario, the crime data were provided in an point-of-occurrence format. This probably is not possible, due to police agencies' concerns about privacy of the people involved in the crime incident. Yet, using in the second part of the analysis process zip code area level units of geography provided much better resolution and solved part of the problem described.
I hope, that this study represents a beginning in the interdisciplinary study of communication and public imagination, which unites state of the art computer technologies, such as GIS, with communication and sociological research, which will be soon followed by scholars from across the field.
Figure 2.
Figure 3.
Figure 4.
Figure 5.
Figure 6.
Figure 7a.
Figure 7b.
REFERENCES:
Bailey, T. C. (1994). A review of statistical spatial analysis in GIS. In A. S. Fotheringham & P. Rogerson (Eds.), Spatial analysis and GIS (pp. 13-44). London: Taylor and Francis.
Ball-Rokeach, S. J. (1985). The origins of individual media system dependency: A sociological framework. Communication Research, 12, 485-510.
Barber, G. M. (1988). Elementary statistics for Geographers. New York: The Guilford Press.
Cassirer, E. (1944). An essay on man. New Haven, CT: Yale University Press.
Carter, R. (1979). The criminal's image of the city. New York: Pergamon Press.
CJSC. (1998). Criminal Justice Statistics Center: Statistics (On-line. Available: http://caag.state.ca.us/cjsc/datatabs.htm ). Sacramento, CA: CJSC.
Cressie, N. A. C. (1993). Statistics for spatial data. New York: John Wiley & Sons, Inc.
DeFrances, C., & Smith, S. K. (1998). Perceptions of Neighborhood Crime, 1995 (Special Report NCJ-165811). Washington, DC: U.S. Department of Justice, Office of Justice Programs.
Gerbner, G. (1971). Violence in television drama: trends and symbolic functions. In G. A. Comstock & E. A. Rubinstein (Eds.), Television and social behavior (Vol. Vol. I, Media content and control, ). Washington, D.C.: U.S. Government Printing Office.
Gerbner, G., Gross, L., Morgan, M., & Signorelli, N. (1994). Growing up with television: The cultivation perspective. In J. Bryant & D. Zillmann (Eds.), Media Effects: Advances in Theory and Research. Hillsdale, N.J.: L. Erlbaum Associates.
Gitlin, T. (1980). The whole world is watching: mass media in the making & unmaking of the New Left. Berkeley, CA: University of California Press.
Griffith, D. A. (1996). Spatial autocorrelation and eigenfunctions of the geographic weights matrix accompanying geo-referenced data. Canadian Geographer, 40(4), 251-367.
Langer, S. (1948). Philosophy in a new key. New York: Penguin Books.
Lynch, K. (1960). The image of the city. Cambridge, MA: MIT Press.
Kaluzny, S. P., Vega, S., Cardoso, T. P., & Shelly, A. A. (1998). S+SpatialStats. User's manual for windows and UNIX. New York: Springer-Verlag.
Kelly, G. A. (1955). The psychology of personal constructs. New York: Norton.
Kitchin, R. M. (1997). Exploring Spatial Thought. Environment and Behavior, 29(1), 123-156.
LAPD. (1998). Los Angeles Police Department Website (On-line. Available: http://www.lapdonline.org/home.htm ). Los Angeles: LAPD.
LASD. (1997). County of Los Angeles, Sheriff's Department: Year in Review (Yearbook ). Los Angeles: County of Los Angeles, Sheriff's Department.
Levine, N. (1996). Spatial statistics and GIS: Software tools to quantify spatial patterns. Journal of American Planning Association, 62(3), 381-389.
Morenoff, J., & Sampson, R. J. (1997). Violent crime and the spatial dynamics of neighborhood transition: Chicago, 1970-1990. Social Forces, 76(1), 31-64.
Myers, S. L., & Chung, C. (1998). Criminal perceptions and violent criminal victimization. Contemporary Economic Policy, 16(3).
Rountree, P. W., & Kenneth, C. L. (1996). Perceived risk versus fear of crime: empirical evidence of conceputally distinct reactions in survey data. Social Forces, 74(4).
Sawicki, D. S. (1996). Neighborhood indicators: A review of literature and an assessment of conceptual and methodological issues. Journal of american Planning Association, 62(2), 165.
Scheufele, D. A. (1999). Framing as a theory of media effects. Journal of Communication, 49(1), 103-122.
Tamaki, J. (1999, January 24). Southland's winners and losers in 1998. The Los Angeles Times, pp. K-1.
Turner, E., & Allan, J. (1989). The most ethnically diverse urban places in the United States. Urban Geography, 10(6), 523-529.
The Los Angeles_Times. (1998, January 18). Winners and losers of 1997. The Los Angeles Times, pp. K-1.