|
|||
|
|
|
|
|
|
By Steven A. Camarota Statistical Findings The tables and maps examined so far provide a good deal of information on immigrant-related poverty. The tables in particular provided detailed information on immigrant-related poverty. However, looking at poverty one characteristic at a time can only tell part of the story because variables are often highly correlated. For example, persons with few years of schooling also tend to have larger families than more educated individuals. It is not possible using a simple tabulation to determine whether the higher poverty rate for high school dropouts is explained by their lack of education or their larger family size. This problem becomes even more acute when many variables are involved, as is the case in this study. A number of statistical methods can be used to examine social phenomena, such as poverty, while controlling for many factors at once. These methods make it possible to estimate the impact of one variable, such as education, while holding other factors constant. Because the variable of interest here can take only two values (in poverty or not in poverty), a logistic regression is performed using the March 1998 CPS. The unit of analysis in the model is the individual. Household level variables such as the race of household head or the number of children in the households are assigned to each person based on the characteristics of the households in which he or she resides21. The model takes the following form:
Individual's poverty statusi = + ß1 X + ß2 Z + e
where X is a vector of household level variables including nativity of head, education of head, size of household and other household characteristics; Z is a vector of individual level variables; and e is an error term. Table 16 provides brief descriptions of each of the variables used in the equation.
Logistic Regression. The results of the logistic regression are presented in Table 17. The model does a reasonablely good job of predicting who is in poverty, predicting over 90 percent of the cases correcting. While this indicates very strong predictive power for the model, one must remember that since the national poverty rate at the time of the survey was 13.3 percent, guessing that an individual is not in poverty for every case (zero in the model) would be correct 86.7 percent of the time. In comparison, the model predicts poverty status correctly 90.3 percent of the time, a modest but not insignificant increase of 3.6 percentage points. Only the Head Female variable is statistically insignificant. All other variables are significant at the .001 level. The log-odds from the logistic regression reported in Table 17 are difficult to interpret. Therefore, it is necessary to calculate predicted probabilities for each variable in the equation to more easily interpret their impact on the propensity of an individual to be in poverty.
Predicted Probabilities When One Characteristic is Varied. One of the benefits of doing a logistic regression analysis is that it is possible to calculate the probability of an individual being in poverty given a certain set of characteristics. This can be done by holding all variables constant except for the variable of interest, and then varying that variable to determine its individual effect on the probability that a person will be in poverty. The probabilities reported in Table 18 are derived by varying only one variable at a time, all other characteristics, including those not reported in the table, are set to their mean value.22 The table can be interpreted as follows: Holding all other variables in the equation constant, an individual in an immigrant household has a .19 probability or 19 percent chance of being in poverty. In comparison, an individual in a native household has a .12 probability of being in poverty, again holding all other variables in the equation constant. This means that even after accounting for a wide variety of factors including, race, education, age, number of children in the household, and family structure, an individual in an immigrant household is about 50 percent more likely to be in poverty than a person in a native household. The predicted probabilities show that the effect of being in an immigrant household is similar in size to other factors often associated with a high risk of poverty. For example, the chance of being in poverty for persons in immigrant households is similar in size to the .21 probability that exists if an individual resides in a household headed by an unmarried female. This is also true for the labor force status of the household head. In the model, the probability of being in poverty for persons in a household with a non-working head is .20, almost the same as the .19 probability for persons in immigrant households. While it has received relatively little attention from researchers, this analysis indicates that being in an immigrant households increases significantly the risk that an individual will be in poverty.
Predicted Probabilities When Two Characteristics Are Varied. So far only one characteristic at a time has been varied; however it is possible to vary more than one variable simultaneously in order to better understand how factors interact with one another. This is potentially important because such factors as education, or marital status might have a different impact on persons in immigrant households than persons in native households. Table 19 reports probabilities when the immigrant variable and one additional variable is varied at the same time. As is the case with Table 18, all other variables are held constant by being set to their mean values. Thus, the table can be interpreted as follows: Individuals in a household headed by a native who is in the labor force have a 11 percent probability of being in poverty. The probability of being in poverty for persons in a household headed by an immigrant in the labor force is 17 percent. If we look at households headed by natives and immigrants not in the labor force, we see that the percent probability of being in poverty jumps significantly to16 percent for persons in native households and jumps even higher, to 25 percent, for persons in immigrant households. This means that the gap between persons in native and immigrant households increases from 6 percentage points when the head is in the labor force to 9 percent points when the head is not in the labor force. This suggests that not holding a job creates a greater risk of poverty among those in immigrant households than individuals in native households. Low levels of educational attainment also create a greater risk of poverty among persons in immigrant households than among those in native households. For example, if we look at poverty for persons in native and immigrant households by education level, we see that the gap between immigrants and natives is 11 percentage points for persons in households headed by dropouts but only 7 percentage points for persons in households headed by an individual with some college. Table 19 also shows that the same general pattern holds depending on the number of children in a household and if the household is headed by an unmarried woman. The more children in the household, the larger the difference between the poverty rates of natives and immigrants. For households headed by unmarried women, the gap is also substantially larger for persons in immigrant households compared to those in native households. The results in Table 19 should not be overstated. While the increase in the probability of being in poverty for persons residing in an immigrant household is significant, the growth in the size of the gap associated with changes in labor force status, education, number of children, and marital status cannot be described as extremely large. However, a pattern does seem to exist. Certain factors do seem to have a more deleterious effect on persons in immigrant households than those in native households, even when other variables are held constant. Overall the logistic regression and predicted probabilities indicate that the problem of immigrant poverty is very complex. Even after controlling for a wide variety of socio-demographic variables, the findings show that the chance of being in poverty for individuals in immigrant household is significantly higher than that of natives. It seems likely that factors such as unfamiliarity with their new country (in particular, its labor market), language barriers, cultural values, discrimination, and other factors lead to high poverty rates for immigrants and their dependents. Overcoming these problems takes time, and many immigrants are simply unable to do so.
|
