Home ] Participants ] Investigators ] Publications ] Methodology ]

 

Journal of Social Service Research, Vol. 1(2), Winter 1977

Copyright 1978 by The Haworth Press. All rights reserved.

 

Ecological Validity of Indicator Data as Predictors of Survey Findings

 

Leonard S. Kogan

Jesse Smith

Shirley Jenkins

 

The initial impetus for this research, begun in 1970, was the national interest in developing indicators to monitor the physical health and the social, emotional, and cognitive functioning of children. This information was to be input for decisions on allocation of resources and planning. To facilitate the process, the research sought a single indicator that would measure "quality of child life" by incorporating both health and welfare measures. Twenty-four variables were selected representing key aspects of child and "contextual" (i.e., family and community) variables. Data were collected for two time periods (1960 and 1970) and three sets of geographical units (the 50 states of the United States plus the District of Columbia, the 62 counties of New York state, and the 62 community districts of New York City). (For a detailed description of the earlier work, see Kogan & Jenkins, 1974.) One major aim of this study was to identify the variables which were highly related in all six data sets and which , therefore, seemed to describe the state of children reliably both over time and across sets of geographical units.

The variables included in these analyses were selected with the aid of a schema that attempted to distinguish attributes and characteristics from resources and services, child from context, and health from general welfare (see Figure 1). A severe restriction on the selection of variables was the necessity that they be available for the three sets of geographical units. Unfortunately, many potentially valuable variables are not readily available for units smaller than states. In addition to the 24 variables in Figure 1, several demographic variables were employed in the analyses to aid in the interpretation of the results. These were: White Population, Average Family Size, Divorced Marrieds, Under 18 Population, and Urbanization. Most of the variables were expressed as rates per unit population. All were subjected to factor analysis.

Of the 24 normative variables, 5 showed high loadings (above a criterion of .60) on the first principal factor in all six independent analyses.1 This first factor was interpreted to represent an underlying dimension associated with poverty and discrimination, since percentage of white population was always very highly but negatively loaded on the first factor and welfare dependency was also highly loaded. "This first factor was called DISORGANIZED POVERTY (describing the negative pole of the factor), and an index that combined the five highly intercorrelated indicators was labeled the DIPOV Index. The letters in DIPOV forma n acronym based on the initial letters of the five indicators: D for Dependency (proportion of children under 18 in families receiving AFDC – Aid to Families with Dependent Children); I for Incomplete Families (proportion of children under 18 not living with both parents); P for Premature Births (rate of infants with birth weight under 2,501 grams per 1,000 live births); O for Out of Wedlock Births (proportion of live births designated out-of-wedlock); and V for Venereal Disease, Juvenile (usually defined in our data as rate of reported cases of primary or secondary syphilis or gonorrhea among persons under age 20 per 100,000 population under age 20).

The DIPOV Index was assumed to be a first approximation for the representation of "quality of child life" for designated geographical areas, which can be characterized and ranked according to their DIPOV Index values, thus measuring relative degree of "needs" and/or "social problems" of the children in each area. Although this can be hypothesized, without testing there is no confidence in the extent to which the DIPOV Index can serve as a surrogate for a larger set of needs and problems. The indicators that compose the DIPOV index are, for the most part, remote from the children’s actual health and behavior, since four out of five characterize family conditions or conditions of birth and parental behavior. Thus, additional study was necessary by means of collection of a broad range of data concerning a representative sample of children. This took the form of an in-depth sample survey of families with children in counties that differ markedly in their DIPOV Indices. (The sample survey study is reported in full in Kogan, Smith, and Jordan, 1976.)

 

Sample Survey Methodology

 

Interview Instruments

To undertake the sample survey, interview schedules were developed for use with the mothers or mother surrogates of sample children. The child items primarily ask about current, age-appropriate behavior of children and generally attempt to obtain descriptions of specific behaviors rather than broad, evaluative judgments from the mother. Some child items, however, are historical, especially in the area of health. A substantial portion of each interview schedule is designed to measure parental behavior and attitude, family background characteristics, and aspects of the social and physical environment. In addition, there are items directly concerned with the DIPOV variables, so that the mother is asked about the family’s welfare status (Dependency), the composition of the household (Incomplete Families), the birth weight of the child and his siblings (Premature Births), the children’s dates of birth and the mother’s marital history (Out-of-Wedlock Births), and the occurrence of venereal disease among family members under age 20 (Juvenile Venereal Disease). Three age-level schedules were developed (age 1, ages 2 – 4, and ages 5 – 10). They were pretested in New York and in Texas in 160 riled interviews, and later revised. (Fieldwork in Texas was done with the help of the Texas Office of Early Childhood Development, Jeannette Watson, director, and David Nesenholtz, director of planning. Pretests in New York were done initially by staff, and later by the National Opinion Research Center, University of Chicago.)

 

Selection of Counties and Sampling Procedures

The DIPOV Indices for the 62 counties of New York State in 1970, 1971, and 1972 were examined with the purpose of selecting 2 counties for study, 1 with a high DIPOV value and 1 with a low value. The 5 counties of New York City were arbitrarily excluded as atypical of the national scene. Albany County was designated on the the DIPOV scale as one of the "worst" counties and Saratoga County as one of the "best" counties, and these were chosen for study by means of the sample survey. For example, in 1970 Albany had 95 premature births (per 1,000) whereas Saratoga had 66. Albany had 5.9 children (per 100) on AFDC, and Saratoga had 1.5. Albany had 10.8 out –of-wedlock births reported (per 100), and Saratoga had 4.2.

Each of the two selected counties was subjected to a form of probability area sampling in order to obtain representative samples of families with at least one child between the ages of 1 and 10 years. The entire sampling process can be viewed as a four-stage sequential procedure.

Primary Sampling Units were created from Enumeration Districts and Block Groups, which re divisions defined by the Census Bureau, and which when taken together comprise the entire area and population of a larger geographic area such as a county. Enumeration Districts are population areas averaging about 250 housing units, and Block Groups are combinations of contiguous blocks having a combined average population of about 1,000.

Population data from the 1970 census for these Primary Sampling Units were updated for 1975 after consultation with local officials, and adjustments were made for new residential construction. After correction, the Primary Sampling Units in each county were stratified by urban-rural status, proportion of white population, and median income, A systematic sample of Primary Sampling Units in each county was then drawn with probability proportional to size (number of households.)

Each selected Primary Sampling Unit was subdivided for a second-stage sample. Segments were constructed in Enumeration Districts by the use of aerial photographs and survey maps, and block divisions within Block Groups were obtained from census publications. Segments or blocks were them selected with probability proportional to size (number of households), and each was surveyed in the field with a proportion of the households being selected systematically according to a predetermined sampling ratio. Address lists were compiled in this process, and interviewers were sent to the selected addresses. Those households with at least one child between the ages of 1 and 10 years were "qualified" for the study, and when possible, an interview was obtained. In each qualified household, the interviewers, by use of a set of prepared tables, randomly selected one child o those in the appropriate age range.

In Albany County 2,252 households were approached, but 1,749 of these did not contain a child in the study population. Completed interviews were obtained from 424 of the 503 qualified households, yielding a response rate of 84.3% In Saratoga County 2,014 households were screened, but 1,376 were not qualified. Completed interviews were obtained from 552 of the 638 qualified households. This resulted in a response rate of 86.5%. The fieldwork for this study required about 6 months, from January to July 1975. The interviewers were hired in Albany; after training, they conducted interviews in both counties, which are contiguous.

 

Overview of Analyses

 

This sequence of studies can be characterized as an attempt to develop a set of quality of child life" indicators, and then to evaluate the ability of these indicators to depict life quality ecologically (i.e., in successively more proximal environments represented by smaller and smaller geographical units). In line with its logical structure, the evaluation takes the form of a distal-to-proximal ecological progression: counties, Primary Sampling Units (PSUs), neighborhoods, and finally families. PSUs had been used in obtaining probability samples of the counties, and the sampling frame made possible the use of census data for PSUs to characterize ecological settings that would be smaller than counties but larger than neighborhoods. The final sample was composed of 98 PSUs, 49 in each county.

A distal-to-proximal progression of ecological settings may be represented statistically by a "hierarchical" multiple regression model (Cohen & Cohen, 1975), which indicates the extent to which measures of the quality of child life can be predicted from county membership, and them successively indicates the added predictability afforded by PSU, neighborhood, and then family variables. This analytic scheme allows the most distal unit (county) to account for as much variability in each child measure as it can, then permits the next most distal unit (PSU) to account for as much of the remaining variability as it can, and finally allows the more proximal units (neighborhood and family) to account for as much of the remaining variability as they can.

The first step in the analysis was to determine whether proxies for the DIPOV variables that were derived from the survey data would provide the same picture of the two counties as was provided by the available-data DIPOV variables. Compared to Saratoga County, Albany County has a higher rate of Dependency, Incomplete Families, Premature Births, Out-of-Wedlock Births, and Juvenile Venereal Disease, and therefore, a DIPOV Index at the unfavorable end of the scale. As a result of the sample survey in these two counties, proxies for the DIPOV variables were created, based on interview responses. If the DIPOV Indices and the component DIPOV indicators based on available data provide an accurate picture of the counties, and if the county samples are representative, we would expect county membership to predict relative status on the DIPOS proxies, thus cross-validating the available-data indicators.

If expectations are confirmed, the next step is to determine the extent to which a large number of variables, describing such factors as the physical health and the cognitive, social, and emotional functioning of children, are predictable from successively more proximal sets of ecological variables. Both steps are accomplished by employing hierarchical multiple regression models. That is, first, county membership, the most distal of our variables, is used to predict, Then, after the variability due to county status is removed, the next most proximal sets of variables, representing PSU characteristics, are used to predict. (Since DIPOV variables were not available below the county level, proxy variables for PSUs were derived from census data. They included three measures highly correlated with DIPOV in a series of tests: Urbanization, i.e., urban-rural status; Percent White; and Median Income.) In the second model, after the variability due to PSU is removed, a :neighborhood" set is entered into the model, and then, in turn, seven successively more proximal "family" sets of variables are entered. Diagrams showing the models for dross-validating available data and for predicting child health and behavior are shown in Figures 2 and 3, respectively

 

Predicting Survey Data from the Indicators

 

In the hierarchical regression model, the statistical strategy consists of testing the incremental variance accounted for by each successive set of variables, using the test for significance of an incremental R2. An examination of the regression coefficients for variables in the set will usually determine which of the variables in the set are responsible for the observed effect, and enable us to interpret the direction and approximate size of effects.

A summary of the results of this regression analysis appears in Table 1. For each of the criterion variables, the proportion of variance accounted for (R2) or the incremental proportion of variance accounted for (deltaR2) by each predictor set is presented. In addition , if the predictor set as a whole is significant, the beta values (standardized regression coefficients) and their signs are noted. If the significant predictor set contains more than one variable, betas re presented for each variable in the set.

To illustrate what this analysis reveals, let us first consider one of the criterion variables, Dependency. Dependency is not predictable from the first predictor set, subject variables This indicates that these is o relationship between, on the one hand, the Age and Sex of the sample child and, on the other hand, the Dependency status of the family. The second predictor set, which is composed of a single variable, county membership, does predict Dependency (deltaR2 = .013, p < .001), and has a positive beta value, indicating that Saratoga has fewer dependent families than Albany. The magnitude of beta (.117) is not large, but we would not expect it to be since there is considerable overlap between the counties (e.g., most of the families in both counties had no welfare income). The available data on Dependency showed that in 1970 the percentage of children on AFDC in Albany County was 5.9%, and in Saratoga County, 1.5%. The survey data, which provided a Dependency proxy (percentage of families with welfare income in 1974), show 10.7% in Albany County and 4.6% in Saratoga County; thus, available data for Dependency have been essentially cross-validated. To achieve cross-validation, the regression coefficients should be significant and have the appropriate sign, but the magnitude of the beta need not be large. Both of the necessary conditions are met in this case.

The third predictor set also predicts Dependency (deltaR2 = .202, p< .001). Of the three variables in the PSU set, Percent White is the strongest predictor (beta - .387). The positive sign indicates that as the census-derived variable, Percent White, increases among the PSUs, Dependency decreases. (This was the case because the coding for Dependency was 1 for Welfare Income and 2 for No Welfare Income.) Median Income, the next strongest predictor in this set, also has a positive sign. The last variable in this set, Urbanization, also predicts Dependency. The sign in this case is negative, because the coding was 1 + Rural, 2 + Urban. A negative sign is interpreted to indicate that the urban PSUs show more Dependency than the rural PSUs. The deltaR2 associated with the PSU set is substantially larger than the deltaR2 associated with county membership. This illustrates a finding that will be repeatedly met in the data – namely, that far more of the criterion variance is account4d for by PSU membership (indexed here by the three demographic variables) than by county membership. In a sense, this pattern arises because PSUs are more homogeneous than counties.

The "Final R," is the last column of Table 1, is the multiple correlation obtained using all three predictor sets. For Dependency, R= .469 (p < .001), and R2 = .220 is the proportion of variance accounted for by this predictor at the county and PSU level. Considering all DIPOV variables together, Table 1 indicates that both indices and three of the five DIPOV components are predictable from county membership. Furthermore, since all the signs are positive, indicating that Saratoga is "better," we can consider that for these variables, although the betas are small, the available data have been cross-validated. This is the expected finding considering the contrasting counties that were chosen for the survey, provided the available data are accurate and the survey samples are representative.

The two criterion variables that are not predictable from county membership are Premature Births and Juvenile Vene4real Disease. Juvenile Venereal Disease was probably not adequately measured for several reasons, perhaps principally because the sample tended to exclude families with teenage children, but also, since many cases of juvenile venereal disease are treated without parental knowledge, the respondent may not have had the information to answer the item correctly. Furthermore, the question is quite sensitive, and some respondents may have chosen not to respond accurately. (Since so few instances of juvenile venereal disease were reported – 9 cases in the entire sample of 976 – an index omitting this component [DIPO] was employed in many of the analyses in preference to the full index [DIPOV]). Premature Births presented a different problem, and probably reflects the fact that although premature births in the two counties differed substantially in 1970, they were almost identical in 1960, the decade in which most of the sample children were born. (Although Premature Births was not a variable that successfully contrasted the two counties in the study, it appeared to be adequately measured and, therefore, was retained in the index.)

 

Predicting Child Health and Behavior and Selected Parent Variables

 

The second set of analyses also uses a single hierarchical multiple regression model for predicting, in this case, each of 101 criterion variables from 28 ecological and family predictor variables. The 28 predictor variables are grouped in 13 sets, ordered from top to bottom in Figure 3. Thus Set I contains the "subject" variables, Age and Sex, and Set XIII contains the :family discipline" variables, Consistency of Punishment and Respondent Strictness. The criterion variables were conceptualized as falling into several broad domains and subdomains, as shown in Table 2.

A complete presentation of the results of the 101 regression analyses would be excessive and forbidding. Instead, one summary table has been prepared to allow an overview of these analyses. Table 3 indicates what percentage of the 101 criterion variables was found to be significantly predicted y each of the 13 predictor sets. For example, Set III (PSU DIPO Index) successfully predicts 28% of the criterion variables at a better than the .01 level, and 44% at a better than the .05 level. In general, it can be seen that all of the sets successfully predict at least a fair percentage of the criterion variables, and some sets predict a very substantial percentage, in spite of the fact that variance is partialled out set by set. Set XII, for example, predicts 29% of the criterion variables (p < .05) even though the variance associated with the 11 preceding sets was removed before Set XII was entered. It should be noted, however, that as in the cross-validation analyses, the betas for the successful predictors are not large. Here they tend to run in the .10 to .30 range. The multiple Rs at the final step in these analyses are generally in the .30 to .40 range. A description of the detailed results of these analyses will be presented only for predictor Sets II and III, county and PSU DIPO Index, since these are the most important distal predictors in this study.

County

Statistically significant county effects were found for 17% of the criteria with typically 1% to 3% of the variance accounted for. These will be interpreted by describing the significant effects from the standpoint of the Saratoga children, who were hypothesized to be healthier, better adjusted, and in general to have fewer problems than the Albany children.

The results show that Saratoga children are less likely to have major disorders with extreme behavioral implications and sleep problems (2 – 4 years of age), and less likely to be "delinquent" (5 – 10). Although there is no difference between the counties in the proportion of children who watch TV, Saratoga children (2 – 20) spend fewer hours per day watching TV. In addition, Saratoga children are temperamentally less intense (5 – 10) and more adaptable (1 – 4), and their mothers rate them higher in arithmetic ability (5 – 10). Thus, on a number of indices scattered through the organismic-behavioral domains, Saratoga and Albany children differ, and in general, the differences favor the Saratoga children. Differences also exist for other criteria, most of them parental. The data also show that Saratoga mothers make use of lay advice (friends and relatives), and the Albany mothers make more use of institutional services in dealing with health and behavior problems of the children.

PSU DIPO INDEX

Families were characterized for DIPO status by counting the number of adverse conditions – welfare status, absence of a legal husband in the home (incomplete family), and prematurity or out-of-wedlock status of one or more children in the family. Then the 98 PSUs were characterized fo4 DIPO status by taking the mean of the families in each PSU, to determine if this PSU DIPO Index proxy would predict the health and functioning of the children.

Of the 101 criteria, 44 show a significant PSU DIPO Index effect, and the effects are overwhelmingly in the predicted direction., Generally, the index accounts for from 1% to 4% of the variance for these 44 successfully predicted criteria. Among the health variables, children from better PSUs are less likely to have had severe measles or mumps, major disorders with extreme behavioral implications, major hospitalizations, eye problems, and possible motor problems. They have had fewer diseases and hospitalizations, and their mothers rate them higher in general physical health. They are more likely to have a good breakfast (5 – 10 years of age) and regular medical caretaking, and are slightly taller.

Among the social-emotional variables, children from better PSUs are less active (1 – 10 years of age), less intense (1 – 4), less irregular in habits (1 –4), less irritable (5 – 10), more distractible (1 – 4), and more adaptable (1 – 4). They are less likely to be difficult children (1 – 4), "internalized" (5 – 10), "self-destructive/noncompliant," "antisocial" (5 – 10), jealous and selfish (2 – 20), but somewhat more likely to be argumentative and show sever mood shifts. They are less likely to have tics, frequent anger, and to have run away from home (5 – 10). The quality of their interaction with other children (5 – 10) and siblings (2 – 10) is better. Mothers in better PSUs tend to use fewer discipline methods (both positive and "strong" negative) when the child misbehaves and, based on responses to two hypothetical situations, use more positive than "strong" negative discipline methods. Parents in better PSUs tend to be more consistent and strict, but are not overprotective (5 – 10). Among the cognitive variables, children I better PSUs have higher general cognitive competence (2 – 10) and arithmetic ability (5 – 10), their mothers have higher educational aspirations (2 – 10) and expectations (5 – 10) for the children, and are more likely to provide cognitive stimulation (2 – 10). What this indicates is that a simple composite of four out of the five DIPOV components, measured at the level of PSUs, is significantly related to a wide variety of normative criterion variables. Thus, the measure of DIPO at the level of PSU is related to many more aspects of child health and welfare than at the level of counties, and the size of the relationships is generally larger.

PROBLEMS OF PREDICTIONS

It should be noted that only 15 of the 101 criterion variables were not related to any of the predictor sets. These are largely from the health category, such as ILLNESS INDEX, Ear Problems, Regular Use of Medicine, Eating Problems (2 – 4), and Headaches, and a few temperament variables from the social-emotional category, such as Persistence (1 – 4) and Quality of Interaction with Other Children (2 – 4). Most of these refer only to the very young child. Another special group are 18 criterion variables that are predictable only from the family predictor sets (VII – XIII). They include Major Pregnancy Problems, Birth Problems (of the child), Major Health Problems, Sleep Problems (5 – 10), Eating Problems (5 – 10), Extreme Mood Shifts (1 – 4), and "Attention-seeking." Only 1 of these variables I cognitive, but otherwise they are scattered through the organismic-behavioral domains and subdomains. However, the 19 parental variables in the study are predictable from one or more of the distal units – county, PSU, and neighborhood. There is, no doubt, more geographical homogeneity among the parents, who have "chosen" where they live, than there is among the children.

 

Discussion

 

In substance, at the county level, the available data can be said to have been cross-validated. The analysis demonstrate that both indices and three of the five DIPOV components are predictable in the appropriate direction from county status. The exceptions are Premature Births and Juvenile Venereal Disease, which re not predictable at the county level. In the case of Juvenile Venereal Disease, this can be attributed partially to lack of measurement of teenagers in the survey and, in the case of Premature Births, to an insufficient difference between the two counties when the entire relevant time span (1964 – 1974) is considered.

 

At the PSU level, the available data are even more consistently cross-validated. All seven of the DIPOV variables (the five components and the two indices) are predictable from the PSU set, even though this does not hold in every instance for each particular predictor variable. Furthermore, on the whole the strength of the association with the DIPOV proxies is greater for the PSU variables than for the county variables.

Although some important child and parent variables are significantly associated with county membership, the total number of variables predicted is relatively small. Also, the strength of the associations is not great. It is not clear why this is so, since there are differences of fair size in Dependency, Incomplete Families, and Out-of-Wedlock Births. The limitation of this study to upstate counties, omitting the more extreme counties in New York City, may have reduced the possibility of substantial findings for counties. Another influential factor may be the restriction of this study to children below the age of 11, since older children ten to show effects of deficits more extremely.

 

 

In contrast to the county results, the PSU Index is significantly associated with a very substantial number of child and parental variables, and the effects in general are somewhat stronger. A considerable proportion of both child and parent variables are successfully predicted, and these belong to all of the organismic-behavioral domains; health, social-emotional, and cognitive. This indicates that, although this predictor does not account for a considerable proportion of variance, it has utility in its significant association with a wide range of variables. Thus, where periodic family surveys are not feasible, DIPOV Indices based on available data can be useful in monitoring a wide spectrum of behavior.

 

 

 

Journal of Social Service Research, Vol. 1(2), Winter 1977

Copyright 1978 by The Haworth Press. All rights reserved.

 

Ecological Validity of Indicator Data as Predictors of Survey Findings

 

Leonard S. Kogan

Jesse Smith

Shirley Jenkins

 

The initial impetus for this research, begun in 1970, was the national interest in developing indicators to monitor the physical health and the social, emotional, and cognitive functioning of children. This information was to be input for decisions on allocation of resources and planning. To facilitate the process, the research sought a single indicator that would measure "quality of child life" by incorporating both health and welfare measures. Twenty-four variables were selected representing key aspects of child and "contextual" (i.e., family and community) variables. Data were collected for two time periods (1960 and 1970) and three sets of geographical units (the 50 states of the United States plus the District of Columbia, the 62 counties of New York state, and the 62 community districts of New York City). (For a detailed description of the earlier work, see Kogan & Jenkins, 1974.) One major aim of this study was to identify the variables which were highly related in all six data sets and which , therefore, seemed to describe the state of children reliably both over time and across sets of geographical units.

The variables included in these analyses were selected with the aid of a schema that attempted to distinguish attributes and characteristics from resources and services, child from context, and health from general welfare (see Figure 1). A severe restriction on the selection of variables was the necessity that they be available for the three sets of geographical units. Unfortunately, many potentially valuable variables are not readily available for units smaller than states. In addition to the 24 variables in Figure 1, several demographic variables were employed in the analyses to aid in the interpretation of the results. These were: White Population, Average Family Size, Divorced Marrieds, Under 18 Population, and Urbanization. Most of the variables were expressed as rates per unit population. All were subjected to factor analysis.

Of the 24 normative variables, 5 showed high loadings (above a criterion of .60) on the first principal factor in all six independent analyses.1 This first factor was interpreted to represent an underlying dimension associated with poverty and discrimination, since percentage of white population was always very highly but negatively loaded on the first factor and welfare dependency was also highly loaded. "This first factor was called DISORGANIZED POVERTY (describing the negative pole of the factor), and an index that combined the five highly intercorrelated indicators was labeled the DIPOV Index. The letters in DIPOV forma n acronym based on the initial letters of the five indicators: D for Dependency (proportion of children under 18 in families receiving AFDC – Aid to Families with Dependent Children); I for Incomplete Families (proportion of children under 18 not living with both parents); P for Premature Births (rate of infants with birth weight under 2,501 grams per 1,000 live births); O for Out of Wedlock Births (proportion of live births designated out-of-wedlock); and V for Venereal Disease, Juvenile (usually defined in our data as rate of reported cases of primary or secondary syphilis or gonorrhea among persons under age 20 per 100,000 population under age 20).

The DIPOV Index was assumed to be a first approximation for the representation of "quality of child life" for designated geographical areas, which can be characterized and ranked according to their DIPOV Index values, thus measuring relative degree of "needs" and/or "social problems" of the children in each area. Although this can be hypothesized, without testing there is no confidence in the extent to which the DIPOV Index can serve as a surrogate for a larger set of needs and problems. The indicators that compose the DIPOV index are, for the most part, remote from the children’s actual health and behavior, since four out of five characterize family conditions or conditions of birth and parental behavior. Thus, additional study was necessary by means of collection of a broad range of data concerning a representative sample of children. This took the form of an in-depth sample survey of families with children in counties that differ markedly in their DIPOV Indices. (The sample survey study is reported in full in Kogan, Smith, and Jordan, 1976.)

 

Sample Survey Methodology

 

Interview Instruments

To undertake the sample survey, interview schedules were developed for use with the mothers or mother surrogates of sample children. The child items primarily ask about current, age-appropriate behavior of children and generally attempt to obtain descriptions of specific behaviors rather than broad, evaluative judgments from the mother. Some child items, however, are historical, especially in the area of health. A substantial portion of each interview schedule is designed to measure parental behavior and attitude, family background characteristics, and aspects of the social and physical environment. In addition, there are items directly concerned with the DIPOV variables, so that the mother is asked about the family’s welfare status (Dependency), the composition of the household (Incomplete Families), the birth weight of the child and his siblings (Premature Births), the children’s dates of birth and the mother’s marital history (Out-of-Wedlock Births), and the occurrence of venereal disease among family members under age 20 (Juvenile Venereal Disease). Three age-level schedules were developed (age 1, ages 2 – 4, and ages 5 – 10). They were pretested in New York and in Texas in 160 riled interviews, and later revised. (Fieldwork in Texas was done with the help of the Texas Office of Early Childhood Development, Jeannette Watson, director, and David Nesenholtz, director of planning. Pretests in New York were done initially by staff, and later by the National Opinion Research Center, University of Chicago.)

 

Selection of Counties and Sampling Procedures

The DIPOV Indices for the 62 counties of New York State in 1970, 1971, and 1972 were examined with the purpose of selecting 2 counties for study, 1 with a high DIPOV value and 1 with a low value. The 5 counties of New York City were arbitrarily excluded as atypical of the national scene. Albany County was designated on the the DIPOV scale as one of the "worst" counties and Saratoga County as one of the "best" counties, and these were chosen for study by means of the sample survey. For example, in 1970 Albany had 95 premature births (per 1,000) whereas Saratoga had 66. Albany had 5.9 children (per 100) on AFDC, and Saratoga had 1.5. Albany had 10.8 out –of-wedlock births reported (per 100), and Saratoga had 4.2.

Each of the two selected counties was subjected to a form of probability area sampling in order to obtain representative samples of families with at least one child between the ages of 1 and 10 years. The entire sampling process can be viewed as a four-stage sequential procedure.

Primary Sampling Units were created from Enumeration Districts and Block Groups, which re divisions defined by the Census Bureau, and which when taken together comprise the entire area and population of a larger geographic area such as a county. Enumeration Districts are population areas averaging about 250 housing units, and Block Groups are combinations of contiguous blocks having a combined average population of about 1,000.

Population data from the 1970 census for these Primary Sampling Units were updated for 1975 after consultation with local officials, and adjustments were made for new residential construction. After correction, the Primary Sampling Units in each county were stratified by urban-rural status, proportion of white population, and median income, A systematic sample of Primary Sampling Units in each county was then drawn with probability proportional to size (number of households.)

Each selected Primary Sampling Unit was subdivided for a second-stage sample. Segments were constructed in Enumeration Districts by the use of aerial photographs and survey maps, and block divisions within Block Groups were obtained from census publications. Segments or blocks were them selected with probability proportional to size (number of households), and each was surveyed in the field with a proportion of the households being selected systematically according to a predetermined sampling ratio. Address lists were compiled in this process, and interviewers were sent to the selected addresses. Those households with at least one child between the ages of 1 and 10 years were "qualified" for the study, and when possible, an interview was obtained. In each qualified household, the interviewers, by use of a set of prepared tables, randomly selected one child o those in the appropriate age range.

In Albany County 2,252 households were approached, but 1,749 of these did not contain a child in the study population. Completed interviews were obtained from 424 of the 503 qualified households, yielding a response rate of 84.3% In Saratoga County 2,014 households were screened, but 1,376 were not qualified. Completed interviews were obtained from 552 of the 638 qualified households. This resulted in a response rate of 86.5%. The fieldwork for this study required about 6 months, from January to July 1975. The interviewers were hired in Albany; after training, they conducted interviews in both counties, which are contiguous.

 

Overview of Analyses

 

This sequence of studies can be characterized as an attempt to develop a set of quality of child life" indicators, and then to evaluate the ability of these indicators to depict life quality ecologically (i.e., in successively more proximal environments represented by smaller and smaller geographical units). In line with its logical structure, the evaluation takes the form of a distal-to-proximal ecological progression: counties, Primary Sampling Units (PSUs), neighborhoods, and finally families. PSUs had been used in obtaining probability samples of the counties, and the sampling frame made possible the use of census data for PSUs to characterize ecological settings that would be smaller than counties but larger than neighborhoods. The final sample was composed of 98 PSUs, 49 in each county.

A distal-to-proximal progression of ecological settings may be represented statistically by a "hierarchical" multiple regression model (Cohen & Cohen, 1975), which indicates the extent to which measures of the quality of child life can be predicted from county membership, and them successively indicates the added predictability afforded by PSU, neighborhood, and then family variables. This analytic scheme allows the most distal unit (county) to account for as much variability in each child measure as it can, then permits the next most distal unit (PSU) to account for as much of the remaining variability as it can, and finally allows the more proximal units (neighborhood and family) to account for as much of the remaining variability as they can.

The first step in the analysis was to determine whether proxies for the DIPOV variables that were derived from the survey data would provide the same picture of the two counties as was provided by the available-data DIPOV variables. Compared to Saratoga County, Albany County has a higher rate of Dependency, Incomplete Families, Premature Births, Out-of-Wedlock Births, and Juvenile Venereal Disease, and therefore, a DIPOV Index at the unfavorable end of the scale. As a result of the sample survey in these two counties, proxies for the DIPOV variables were created, based on interview responses. If the DIPOV Indices and the component DIPOV indicators based on available data provide an accurate picture of the counties, and if the county samples are representative, we would expect county membership to predict relative status on the DIPOS proxies, thus cross-validating the available-data indicators.

If expectations are confirmed, the next step is to determine the extent to which a large number of variables, describing such factors as the physical health and the cognitive, social, and emotional functioning of children, are predictable from successively more proximal sets of ecological variables. Both steps are accomplished by employing hierarchical multiple regression models. That is, first, county membership, the most distal of our variables, is used to predict, Then, after the variability due to county status is removed, the next most proximal sets of variables, representing PSU characteristics, are used to predict. (Since DIPOV variables were not available below the county level, proxy variables for PSUs were derived from census data. They included three measures highly correlated with DIPOV in a series of tests: Urbanization, i.e., urban-rural status; Percent White; and Median Income.) In the second model, after the variability due to PSU is removed, a :neighborhood" set is entered into the model, and then, in turn, seven successively more proximal "family" sets of variables are entered. Diagrams showing the models for dross-validating available data and for predicting child health and behavior are shown in Figures 2 and 3, respectively

 

 

Predicting Survey Data from the Indicators

 

In the hierarchical regression model, the statistical strategy consists of testing the incremental variance accounted for by each successive set of variables, using the test for significance of an incremental R2. An examination of the regression coefficients for variables in the set will usually determine which of the variables in the set are responsible for the observed effect, and enable us to interpret the direction and approximate size of effects.

A summary of the results of this regression analysis appears in Table 1. For each of the criterion variables, the proportion of variance accounted for (R2) or the incremental proportion of variance accounted for (deltaR2) by each predictor set is presented. In addition , if the predictor set as a whole is significant, the beta values (standardized regression coefficients) and their signs are noted. If the significant predictor set contains more than one variable, betas re presented for each variable in the set.

To illustrate what this analysis reveals, let us first consider one of the criterion variables, Dependency. Dependency is not predictable from the first predictor set, subject variables This indicates that these is o relationship between, on the one hand, the Age and Sex of the sample child and, on the other hand, the Dependency status of the family. The second predictor set, which is composed of a single variable, county membership, does predict Dependency (deltaR2 = .013, p < .001), and has a positive beta value, indicating that Saratoga has fewer dependent families than Albany. The magnitude of beta (.117) is not large, but we would not expect it to be since there is considerable overlap between the counties (e.g., most of the families in both counties had no welfare income). The available data on Dependency showed that in 1970 the percentage of children on AFDC in Albany County was 5.9%, and in Saratoga County, 1.5%. The survey data, which provided a Dependency proxy (percentage of families with welfare income in 1974), show 10.7% in Albany County and 4.6% in Saratoga County; thus, available data for Dependency have been essentially cross-validated. To achieve cross-validation, the regression coefficients should be significant and have the appropriate sign, but the magnitude of the beta need not be large. Both of the necessary conditions are met in this case.

The third predictor set also predicts Dependency (deltaR2 = .202, p< .001). Of the three variables in the PSU set, Percent White is the strongest predictor (beta - .387). The positive sign indicates that as the census-derived variable, Percent White, increases among the PSUs, Dependency decreases. (This was the case because the coding for Dependency was 1 for Welfare Income and 2 for No Welfare Income.) Median Income, the next strongest predictor in this set, also has a positive sign. The last variable in this set, Urbanization, also predicts Dependency. The sign in this case is negative, because the coding was 1 + Rural, 2 + Urban. A negative sign is interpreted to indicate that the urban PSUs show more Dependency than the rural PSUs. The deltaR2 associated with the PSU set is substantially larger than the deltaR2 associated with county membership. This illustrates a finding that will be repeatedly met in the data – namely, that far more of the criterion variance is account4d for by PSU membership (indexed here by the three demographic variables) than by county membership. In a sense, this pattern arises because PSUs are more homogeneous than counties.

The "Final R," is the last column of Table 1, is the multiple correlation obtained using all three predictor sets. For Dependency, R= .469 (p < .001), and R2 = .220 is the proportion of variance accounted for by this predictor at the county and PSU level. Considering all DIPOV variables together, Table 1 indicates that both indices and three of the five DIPOV components are predictable from county membership. Furthermore, since all the signs are positive, indicating that Saratoga is "better," we can consider that for these variables, although the betas are small, the available data have been cross-validated. This is the expected finding considering the contrasting counties that were chosen for the survey, provided the available data are accurate and the survey samples are representative.

The two criterion variables that are not predictable from county membership are Premature Births and Juvenile Vene4real Disease. Juvenile Venereal Disease was probably not adequately measured for several reasons, perhaps principally because the sample tended to exclude families with teenage children, but also, since many cases of juvenile venereal disease are treated without parental knowledge, the respondent may not have had the information to answer the item correctly. Furthermore, the question is quite sensitive, and some respondents may have chosen not to respond accurately. (Since so few instances of juvenile venereal disease were reported – 9 cases in the entire sample of 976 – an index omitting this component [DIPO] was employed in many of the analyses in preference to the full index [DIPOV]). Premature Births presented a different problem, and probably reflects the fact that although premature births in the two counties differed substantially in 1970, they were almost identical in 1960, the decade in which most of the sample children were born. (Although Premature Births was not a variable that successfully contrasted the two counties in the study, it appeared to be adequately measured and, therefore, was retained in the index.)

 

Predicting Child Health and Behavior and Selected Parent Variables

 

The second set of analyses also uses a single hierarchical multiple regression model for predicting, in this case, each of 101 criterion variables from 28 ecological and family predictor variables. The 28 predictor variables are grouped in 13 sets, ordered from top to bottom in Figure 3. Thus Set I contains the "subject" variables, Age and Sex, and Set XIII contains the :family discipline" variables, Consistency of Punishment and Respondent Strictness. The criterion variables were conceptualized as falling into several broad domains and subdomains, as shown in Table 2.

A complete presentation of the results of the 101 regression analyses would be excessive and forbidding. Instead, one summary table has been prepared to allow an overview of these analyses. Table 3 indicates what percentage of the 101 criterion variables was found to be significantly predicted y each of the 13 predictor sets. For example, Set III (PSU DIPO Index) successfully predicts 28% of the criterion variables at a better than the .01 level, and 44% at a better than the .05 level. In general, it can be seen that all of the sets successfully predict at least a fair percentage of the criterion variables, and some sets predict a very substantial percentage, in spite of the fact that variance is partialled out set by set. Set XII, for example, predicts 29% of the criterion variables (p < .05) even though the variance associated with the 11 preceding sets was removed before Set XII was entered. It should be noted, however, that as in the cross-validation analyses, the betas for the successful predictors are not large. Here they tend to run in the .10 to .30 range. The multiple Rs at the final step in these analyses are generally in the .30 to .40 range. A description of the detailed results of these analyses will be presented only for predictor Sets II and III, county and PSU DIPO Index, since these are the most important distal predictors in this study.

County

Statistically significant county effects were found for 17% of the criteria with typically 1% to 3% of the variance accounted for. These will be interpreted by describing the significant effects from the standpoint of the Saratoga children, who were hypothesized to be healthier, better adjusted, and in general to have fewer problems than the Albany children.

The results show that Saratoga children are less likely to have major disorders with extreme behavioral implications and sleep problems (2 – 4 years of age), and less likely to be "delinquent" (5 – 10). Although there is no difference between the counties in the proportion of children who watch TV, Saratoga children (2 – 20) spend fewer hours per day watching TV. In addition, Saratoga children are temperamentally less intense (5 – 10) and more adaptable (1 – 4), and their mothers rate them higher in arithmetic ability (5 – 10). Thus, on a number of indices scattered through the organismic-behavioral domains, Saratoga and Albany children differ, and in general, the differences favor the Saratoga children. Differences also exist for other criteria, most of them parental. The data also show that Saratoga mothers make use of lay advice (friends and relatives), and the Albany mothers make more use of institutional services in dealing with health and behavior problems of the children.

PSU DIPO INDEX

Families were characterized for DIPO status by counting the number of adverse conditions – welfare status, absence of a legal husband in the home (incomplete family), and prematurity or out-of-wedlock status of one or more children in the family. Then the 98 PSUs were characterized fo4 DIPO status by taking the mean of the families in each PSU, to determine if this PSU DIPO Index proxy would predict the health and functioning of the children.

Of the 101 criteria, 44 show a significant PSU DIPO Index effect, and the effects are overwhelmingly in the predicted direction., Generally, the index accounts for from 1% to 4% of the variance for these 44 successfully predicted criteria. Among the health variables, children from better PSUs are less likely to have had severe measles or mumps, major disorders with extreme behavioral implications, major hospitalizations, eye problems, and possible motor problems. They have had fewer diseases and hospitalizations, and their mothers rate them higher in general physical health. They are more likely to have a good breakfast (5 – 10 years of age) and regular medical caretaking, and are slightly taller.

Among the social-emotional variables, children from better PSUs are less active (1 – 10 years of age), less intense (1 – 4), less irregular in habits (1 –4), less irritable (5 – 10), more distractible (1 – 4), and more adaptable (1 – 4). They are less likely to be difficult children (1 – 4), "internalized" (5 – 10), "self-destructive/noncompliant," "antisocial" (5 – 10), jealous and selfish (2 – 20), but somewhat more likely to be argumentative and show sever mood shifts. They are less likely to have tics, frequent anger, and to have run away from home (5 – 10). The quality of their interaction with other children (5 – 10) and siblings (2 – 10) is better. Mothers in better PSUs tend to use fewer discipline methods (both positive and "strong" negative) when the child misbehaves and, based on responses to two hypothetical situations, use more positive than "strong" negative discipline methods. Parents in better PSUs tend to be more consistent and strict, but are not overprotective (5 – 10). Among the cognitive variables, children I better PSUs have higher general cognitive competence (2 – 10) and arithmetic ability (5 – 10), their mothers have higher educational aspirations (2 – 10) and expectations (5 – 10) for the children, and are more likely to provide cognitive stimulation (2 – 10). What this indicates is that a simple composite of four out of the five DIPOV components, measured at the level of PSUs, is significantly related to a wide variety of normative criterion variables. Thus, the measure of DIPO at the level of PSU is related to many more aspects of child health and welfare than at the level of counties, and the size of the relationships is generally larger.

PROBLEMS OF PREDICTIONS

It should be noted that only 15 of the 101 criterion variables were not related to any of the predictor sets. These are largely from the health category, such as ILLNESS INDEX, Ear Problems, Regular Use of Medicine, Eating Problems (2 – 4), and Headaches, and a few temperament variables from the social-emotional category, such as Persistence (1 – 4) and Quality of Interaction with Other Children (2 – 4). Most of these refer only to the very young child. Another special group are 18 criterion variables that are predictable only from the family predictor sets (VII – XIII). They include Major Pregnancy Problems, Birth Problems (of the child), Major Health Problems, Sleep Problems (5 – 10), Eating Problems (5 – 10), Extreme Mood Shifts (1 – 4), and "Attention-seeking." Only 1 of these variables I cognitive, but otherwise they are scattered through the organismic-behavioral domains and subdomains. However, the 19 parental variables in the study are predictable from one or more of the distal units – county, PSU, and neighborhood. There is, no doubt, more geographical homogeneity among the parents, who have "chosen" where they live, than there is among the children.

 

Discussion

In substance, at the county level, the available data can be said to have been cross-validated. The analysis demonstrate that both indices and three of the five DIPOV components are predictable in the appropriate direction from county status. The exceptions are Premature Births and Juvenile Venereal Disease, which re not predictable at the county level. In the case of Juvenile Venereal Disease, this can be attributed partially to lack of measurement of teenagers in the survey and, in the case of Premature Births, to an insufficient difference between the two counties when the entire relevant time span (1964 – 1974) is considered.

At the PSU level, the available data are even more consistently cross-validated. All seven of the DIPOV variables (the five components and the two indices) are predictable from the PSU set, even though this does not hold in every instance for each particular predictor variable. Furthermore, on the whole the strength of the association with the DIPOV proxies is greater for the PSU variables than for the county variables.

Although some important child and parent variables are significantly associated with county membership, the total number of variables predicted is relatively small. Also, the strength of the associations is not great. It is not clear why this is so, since there are differences of fair size in Dependency, Incomplete Families, and Out-of-Wedlock Births. The limitation of this study to upstate counties, omitting the more extreme counties in New York City, may have reduced the possibility of substantial findings for counties. Another influential factor may be the restriction of this study to children below the age of 11, since older children ten to show effects of deficits more extremely.

In contrast to the county results, the PSU Index is significantly associated with a very substantial number of child and parental variables, and the effects in general are somewhat stronger. A considerable proportion of both child and parent variables are successfully predicted, and these belong to all of the organismic-behavioral domains; health, social-emotional, and cognitive. This indicates that, although this predictor does not account for a considerable proportion of variance, it has utility in its significant association with a wide range of variables. Thus, where periodic family surveys are not feasible, DIPOV Indices based on available data can be useful in monitoring a wide spectrum of behavior.

 

References

Buhler, R. P-Stat: A computing system for file manipulation and statistical analysis of social science data. Princeton: Princeton University Computation Center, 1974.

Cohen, J. & Cohen, P. Applied multiple regression/correlation analysis for the behavioral sciences. Hillsdale, N.J. :Lawrence Erlbaum Associates, 1975.

Harman, H. Modern factor analysis. Chicago: University of Chicago Press, 1967.

Kogan, L.S., & Jenkins, S. Indicators of child health and welfare:Development of the DIPOV Index. New York: Columbia University Press, 1974.

Kogan, L.S., Smith, J., & Jordan, L.A. Children and their families in two counties of New York State: An exploration of the ecological utility of the DIPOV Index. New York: Center for Social Research, City University of New York, April 1976.

 

TABLE 2 Types of Criterion Variables

Category

No. of Variables

Health

Prenatal, perinatal

2

History

12

Present condition

12

Parental care

5

Social-Emotional

Temperament scales

13

Temperament types

5

Indices and traits

29

Parental discipline

8

Cognitive

Child ability

9

Parental-institutional support

6

TOTAL

101