This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Some of the contents of this chapter including many of the tables have appeared in Wagle (2008).
- 2.
The UK government, for example, has shown commitment to systematically curtail social exclusion by adopting definitive policy measures (Davies 2005; Lister 2004; Social Exclusion Unit 2001). This has also been common in the rest of Europe (Glennerster 2002; Littlewood 1999; Mayes et al. 2001) as well as Canada (Crawford 2003; Toye and Infanti 2004).
- 3.
More specifically, variables with missing values included condition of health, treated with respect, occupational prestige, political activism, associational activity, personal contact, and participation in social activities. The imputation of data is justified since all of the important socio-demographic variables have complete data. The variables used in making such predictions include age, gender, nativity, race, marital status, household size, number of adults, number of children, number of earners, education, income, region, dwelling type and ownership, and occupation. While not all variables would turn out to be significant in all cases, I use a consistent set of predictors as it would not lead to more or less biased predictions. Because the original set of the indicators being imputed contains discrete values, I recode the imputed values to gain consistency.
- 4.
The preliminary versions of the model included different specifications in terms of the y vector not only including the content but the form of the Y variables and even more importantly the form of the B matrix. Partly how these matrices are to be conceived depends on the theoretical relationships. But the form and content of the y vector needs to be supported by the given data. This is even more so in case of the B matrix since the literature does not provide specific guidance in terms of these relationships. It was, therefore, important to seek empirical support thus having to test hypotheses related to multiple specifications. This final model, therefore, represents the best possible specification given the data after multiple estimation, respecification, and evaluation attempts.
- 5.
This is to note that the structural equation modeling does not require these indicators to be correlated. I incorporated their correlations for empirical purposes especially since they appeared to be highly correlated, thus considerably improving the model fit.
- 6.
This final model is identified using the t- and two-step rules (Bollen 1989). While I could manually establish identification using these rules, the use of standard software automatically does so in an attempt to estimate SEM models and reports any identification problem. The MPlus software used here indicated that this final version of the model was in fact identified with 109 degrees of freedom.
- 7.
The ratio of Chi square to degrees of freedom, a common indicator of model fit, of 13.64 reported for the model attenuates sizably when the sample size is reduced from the existing 2803 to 25 percent (or 700) randomly selected observations. The ratio of 5.11 for a model with reduced sample size, which is within the commonly acceptable range, conforms to that the overall measure of model fit may not always be reliable. Despite this, however, I continue using the full sample, rather than a smaller, experimental sample, with the conviction that the estimates produced would be more accurate.
- 8.
As happens with most econometric models, taking natural log has produced more robust estimates with both income variables. Also, family income has been equivalized to more appropriately accommodate the effects of family size on sharing income due to economies of scale. Consistent with Citro and Michael (1995) and Short (2001), I equivalized the family income of each respondent using: \({{Y_i } \over {(X_i )^{{1 \over 2}} }}\), where Y is the family income and X is the family size.
- 9.
It must be noted that the R-squared estimates of occupational prestige and employment industry do not accurately mirror the level of their influence in measuring the associated poverty dimension as these load on both capability and economic inclusion.
- 10.
Total effects represent the change in ηi associated to a unit change in ηj. See footnote 16 in Chapter 4 for details.
- 11.
While the process has been purely empirical, this result is not consistent with the case from Kathmandu. Partly it may be a reflection of the contextual dissimilarity as to what extent and how qualitatively one participates in the labor market may not have any systematic effects on the level of material resources, capability, political participation, or civic life. One can be reasonably skeptic, however, as the data captured may not have been complete (my suspicion) or may have behaved differently.
- 12.
As elaborated in Chapter 4 (footnote 17), this transformation was done without changing the relative distribution of the scores.
- 13.
This is assuming that each of the dimensions has equal weight. In reality, people may value any of the economic, political, and civic and cultural inclusions more dearly than others, as it may be more relevant to determining one's social quality of life.
- 14.
The lower and upper bounds that the poverty estimates for the United States are exactly the same as those for Kathmandu. While this makes the comparative analysis simpler, this has resulted from a systematic, not an arbitrary, process.
- 15.
Using census data, Danziger and Gotschalk (2005) estimated the population below official poverty line to be close to 10 percent in 1999. While more recent approaches experimented by the Census Bureau (2006) provide a variety of measurement estimates including those that are as low as slightly over eight percent, these are yet to be formalized as the official poverty lines to be used for a variety of governmental purposes.
- 16.
These are non-cumulative percents indicating that the poverty population under the 10 percent target would exceed 17 percent including all three categories of the poor.
- 17.
This is the case between the absolute 10 percent target and the relative criterion. The absolute 30 percent target provides much larger poverty estimates. There exist some differences even between the former two criteria, however, especially in terms of the American Indians, Hispanics, and two children under six.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2008 Springer Science+Business Media, LLC
About this chapter
Cite this chapter
Wagle, U. (2008). Chapter 5 Application II: The United States. In: Multidimensional Poverty Measurement. Economic Studies in Inequality, Social Exclusion and Well-Being, vol 4. Springer, New York, NY. https://doi.org/10.1007/978-0-387-75875-6_5
Download citation
DOI: https://doi.org/10.1007/978-0-387-75875-6_5
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-0-387-75874-9
Online ISBN: 978-0-387-75875-6
eBook Packages: Business and EconomicsEconomics and Finance (R0)