Journal of Econometrics 43 (1990) 121-133. North-Holland GENERALIZED ENTROPY MEASURES OF MOBILITY DIFFERENT SEXES AND INCOME LEVELS Esfandiar Southern Methodist FOR MAASOUMI University, Dallm, TX 75275, USA Sourushe ZANDVAKILI University of Cincinnati, Cincinnati, OH 45221, USA A family of mobility measures is investigated using the Michigan panel data on income dynamics, These measures are decomposed in order to learn about components that are due to differences in gender and income groups, on the one hand, and within-group components which are free of such group characteristics. These measures compute the degree of equalization over time of the distribution of ‘permanent incomes’. Permanent or average income of an individual is computed over an increasing number of points within a time interval of interest. This degree of equality is compared (as a ratio) to the equality of incomes at any reference point of interest in order to give us a measure of stability in the size distribution of incomes. We take an average of inequality values over the time interval as our reference value for ‘short-run’ inequality. Mobility is greater the greater this degree of relative equalization. Several aggregator functions have been used to compute the ‘permanent income’ variable. Their justification and role in robustifying our inferences is briefly discussed. The mobility profiles here suggest that there has not been much equalization between different sexes, but there is not a lot of inequality between these groups in our sample. Most of the inequality is within groups where the annual distributional changes belie the relative stability of the permanent income distribution. 1. Introduction Mobility may be defined as a measure of movements within the distribution of a suitable welfare attribute such as income. The measurement of such an inevitably latent concept requires a suitable point of reference, or a reference distribution, and entails many implicit value judgements that may be ideally explicated with reference to implied welfare functions. Such measurement should also correct for at least some transitory movements in order to provide meaningful assessments of permanent mobility. Several approaches are available for the study of mobility. The alternatives not discussed in this paper generally require a generation mechanism or an estimate for the transition probabilities governing individual movements. Measures of mobility are functions defined on such probability matrices. For modern welfare-theoretic discussions of this approach see, for example, Shorrocks (1976) and Geweke et al. (1986) in the continuous time domain. 0304-4076/90/$3.5001990, Elsevier Science Publishers B.V. (North-Holland) 122 E. Maasoumi, and S. Zandvakili, Generalized entropy measures of mobility There are other ‘structural’ approaches to earnings mobility that utilize econometric analyses of a priori specified models relating mobility to its hypothesized causes. Examples of this approach are Creedy (1985) and Lillard and Willis (1978). The approach that is implemented here has been developed and applied in Maasoumi and Zandvakili (1986,1988) and Zandvakili (1987). Shorrocks (1978) proposed a special member of the family of mobility measures developed by us, but contains an excellent discussion of the basic ideas. Our measures of mobility are anonymous in the sense that they are based on relative and symmetric measures of ‘distance’ between an income distribution and a uniform (rectangular) one. This is of course a relative inequality measure as exemplified by Theil’s information measures. We also use ‘longer-run’ or ‘permanent’ real incomes by aggregating individual’s income over time intervals of interest. This is designed to control for some of the transitory fluctuations that may be caused by (e.g.) windfalls and to account for other temporary or cyclical movements. But interest usually centers on how the long-run distributions compare with reference distributions at specific points in time. We take the weighted averages of annual inequalities within an interval of interest to be a representation of the ‘short-run’ income inequality and a reasonable reference value. The ratio of these two inequality indices serves as a bounded measure of ‘stability’ in the income distribution over the corresponding time interval. A negative function of this ratio is then a measure of anonymous mobility. Mobility is thus dejned as a relative degree of equalization as the time interval for measuring incomes is expanded. In our investigation we compute permanent incomes using an ‘optimal’ family of aggregator functions proposed in Maasoumi (1986). We also look at diverse members of an ‘optimal’ decomposable family of relative inequality measures known as the Generalized Entropy (GE). The experimentation over members of these families, as well as with different sets of weights given to incomes at different points within the intervals of interest, represent an attempt to robustify our summary findings. We are also able to investigate, in some cases quite unambiguously, decompositions of inequality and mobility profiles by gender and by income level. This level of detail is necessary for a better understanding of where policy may be required. On the basis of the approximately 2300 households which remained in the Michigan panel over the period 1969-1981, we conclude: (i) There is not a great deal of inequality between the men- and women-headed households. (ii) The dominant within-group component of inequality is either increasing over this period or, when incomes are smoothed by time aggregation, relatively stable. (iii) This larger within-group component of inequality is due to high levels of inequality within lower income groups (such as women-headed households). (iv) Grouping by real income brackets leads predictably to very large between-group inequality values. (v) Some equalization of real incomes has occurred over time within most income groups, but this is very hard to E. Maasoumi, and S. Zandvakili, Generalized entropy measures of mobility 123 judge by a comparison of annual inequality measures and most clearly revealed by using our ‘permanent income’ distributions. (vi) Modest levels of mobility are recorded as the aggregation interval is expanded, but the corresponding profiles flatten- out after about eight or nine years. The plan of this paper is as follows. Section 2 briefly outlines the theoretical framework and the notation. Section 3 contains a discussion of our empirical findings with decompositions based on gender. Section 4 deals with similar decompositions based on income groups, and section 5 concludes. 2. The framework Let q, denote the income of individual i, i = 1,. . . , N, in period t, t = 1, . . . , T, and q = cf”_ ,Y,, the total income over any subperiod M I T. A class of aggregate or ‘permanent’ income functions which can treat incomes in different periods as distinct and substitutable attributes may be denoted by M=l,...,T. si=sj(~r,...J,+f)Y (1) Following the arguments in Maasoumi (1986), we choose S,(e) according to generalized information criteria which insure that the distribution of the resulting aggregate income shares has the minimum ‘distance’ from the M distributions of its M annual income components. Examples of such functional families are 1 -l/8 S,a [ EC&~ , f P+0, -1, (2) P=O, (2’) p= -1, (2”) with (Y, denoting the weight given to income at time t, c,a, = 1, and the constant elasticity of substitution of income across time is u = l/(1 + /3). The family of measures used here to compute inequality (the degree of equalization compared with the rectangular distribution) is the Generalized Entropy (GE). For an axiomatic analysis of the many useful properties of these measures see, for instance, Shorrocks (1984). This family includes many of the well-known measures, such as both of Theil’s information indices [see Theil (1967)] and monotonic transformations of a class of measures proposed by Atkinson (1970). The GE family is defined as follows (e.g.) for the 124 E. Maasoumi, and S. Zanduakili, Generalized entropy measures of mobility aggregate income shares, S,* = Si/cy_ Sj: I,(S) = ?[(NS,*)‘+‘- = l]/Ny(l+y), fs;* = &Vi log(Nsi*), log(l/Ns;*), Y#O,-I. Y= 0, y= -1, (3) (3’) (3”) where (3’) and (3”) may be recognized as, respectively, Theil’s first and second inequality measures. The well-known utilitarian, equality-prefering, Schurconcave welfare functions corresponding to this family of indices reveal some of the subjective but traditionally implicit value judgements attached to measuring mobility. Roughly speaking, these welfare functions incorporate a subjective preference for mobility toward an equal distribution of income. Useful additive decomposition properties of these measures have been given elsewhere, for example in Maasoumi and Zandvakili (1988). Accordingly, we are able to distinguish the contribution to total inequality-mobility coming from (averaged) within-groups from that which exists between groups. The least ambiguous of these decompositions is afforded by Theil’s second measure (y= -1). To measure mobility, let Y = (Yi,, ...,Y,,)be the income vector at point t. Then, the measure of mobility described in section 1 is any negative function of the following measure of relative stability, R,, defined over M periods: where the sum in the denominator is a measure of ‘short-run’ inequality and S, are defined over the same M periods in a consistent manner. It is clear that for any convex measure 1, we have 0 I R, I 1 if the permanent income function, S, is a weighted average of incomes with the same weights (Y,[as in (2”)]. This special case was shown by Shorrocks (1978). The same bounds on R, may be established using the decompositions of multidimensional measures of inequality given in Maasoumi (1986, propositions 1 and 2). We may use 1 - R, as a measure of mobility. P, = (R,, ...,RT) is a stability profile which can also reveal the effect of increasing smoothing of the income variable starting from R, = 1. Other than this smoothing out of the short-run effects, R, is capable of revealing durable ‘mobility’ toward equalization in a way that may be obscured by looking at each I,( Y,), t = 1,. . . , T.This is most clearly seen by considering a situation in which only a permutation of the income vector has occurred between two periods. As is well known, this leaves our anonymous E. Muusoumi, and S, Zanduukili, Generulized entropy meusures ofmobility 125 (symmetric), relative inequality measures unchanged, but the distribution of the aggregated incomes over the two periods will be changed, possibly dramatically (unless there is perfect equality to begin with!). Consequently, I,(S) and R M will measure any mobility over the two periods. 3. Mobility and gender ‘Household’ income data for the period 1969-1981 are taken from the Michigan Panel Study. Household’s income (head and spouse, if any) consists of the following: income from wages, salaries, rents, dividends, interest, business, bonuses, commissions, professional practice, aid to dependent children, social security, retirement pay, pension or annuities, unemployment compensation, child support, and other transfer payments. Real total income is obtained using the current consumer price index. Income is adjusted for family size (in 1975) to provide a better measure of family income since family members effectively pool their incomes [see Kakwani (1984) and Rosen (1984)]. We refer to this adjusted income as the ‘Per Capita Family Income’ (PCFI). In computing the permanent incomes three different schemes were used in order to weight income at different times. These (Y, weights are (i) equal weights for all years, (ii) the ratio of mean income at time t to the mean income over the entire M periods (MIW in tables), and (iii) the normalized elements of the eigenvector corresponding to the first principal component of the x’x matrix [see Maasoumi (1989) and Ram (1982)]. We did not find any qualitative differences in our results between these three cases, and thus report only the computations based on ratio of means weights. The other two cases are reported in Zandvakili (1987). In our computations the substitution parameter /? is restricted by the relations - y = 1 + p. We computed four different aggregator functions corresponding to four inequality measures with - y = V= (2,1,0.5,0.0). V= 0.0 and 1.0 correspond to Theil’s first and second inequality measures, respectively, combined with the linear and the Cobb-Douglas forms of the aggregator function. Tables l-3 provide, respectively, the annual short-run inequalities, the inequalities in the aggregated (long-run) incomes, and the income stability of each based on gender is also given. Note that measures R M.Decomposition as one moves toward 1981 the number of periods over which S,, Z,(S), and R, are calculated is increasing from 1 to 13. The results for every other year are reported to save space. Short-run inequality in table 1 has generally increased. As expected, inequality is greater with larger degrees of relative inequality aversion (V). There are 1776 male- and 529 female-headed households in the sample. The ‘withingroup’ component of short-run inequalities is dominant. The absolute value of the ‘between-group’ component, however, has increased over the 13 years. For both sexes annual inequalities have a rising trend (less uniformly so for V = 2). 126 E. Muasoumi, and S. Zandvakili, Generalized entropy measures of mobility Table 1 1969-81 per capita family income: Short-run inequalities. Overall Between Within Men Women 0.793 0.729 0.636 0.815 0.981 1.211 1.520 1.676 1.604 1.600 3.015 2.195 2.580 3.292 Degree of inequality aversion = 2.0 1969 1971 1973 1975 1977 1979 1981 1.109 1.049 1.002 1.636 1.569 1.895 2.441 0.014 0.016 0.019 0.023 0.038 0.041 0.047 1.095 1.033 0.982 1.613 1.530 1.854 2.394 - Degree of inequality aversion = 1.0 1969 1971 1973 1975 1977 1979 1981 0.430 0.466 0.464 0.531 0.578 0.613 0.706 0.013 0.014 0.017 0.021 0.033 0.035 0.039 0.417 0.452 0.446 0.510 0.545 0.579 0.667 0.340 0.375 0.362 0.422 0.467 0.498 0.578 - 0.676 0.709 0.727 0.808 0.808 0.849 0.964 Degree of inequality aversion = 0.5 1969 1971 1973 1975 1977 1979 1981 0.375 0.407 0.404 0.456 0.494 0.505 0.571 0.012 0.014 0.017 0.019 0.030 0.032 0.036 0.363 0.394 0.387 0.436 0.463 0.473 0.535 0.306 0.341 0.331 0.380 0.419 0.427 0.490 - 0.601 0.618 0.635 0.693 0.688 0.707 0.782 Degree of inequality aversion = 0.0 1969 1971 1973 1975 1977 1979 1981 0.367 0.402 0.395 0.448 0.492 0.483 0.549 0.012 0.013 0.016 0.018 0.029 0.030 0.034 0.355 0.388 0.380 0.430 0.463 0.453 0.515 0.304 0.344 0.332 0.383 0.429 0.417 0.481 0.606 0.615 0.631 0.693 0.682 0.688 0.753 But short-run inequality amongst female-headed households is always greater than amongst men. These annual values, however, contain many transitory components which are partially removed from the aggregated values in table 2. Table 2 values exhibit much less volatility. After a decline in the initial years, Z,(S) has increased back to about its original value. Also, long-run inequality is always smaller than the corresponding short-run inequality. Once again, inequality among women is greater than among men, and within-group inequality is several times the between-group component. These relative values are somewhat sensitive to the family size adjustment of incomes. For instance, E. Maasoumi, and S. Zandvakili, Generalized entropy measures of mobility 127 Table 2 1969-1981 per capita family income (MIW): Long-run inequality. Overall Between Within Men Women 0.793 0.665 0.599 0.588 0.602 0.653 0.690 1.616 1.430 1.365 1.533 1.507 1.509 1.563 0.340 0.325 0.316 0.319 0.327 0.336 0.352 0.676 0.651 0.648 0.654 0.662 0.654 0.674 0.306 0.296 0.291 0.294 0.302 0.310 0.323 0.601 0.579 0.577 0.579 0.585 0.579 0.591 0.304 0.295 0.292 0.295 0.306 0.312 0.326 0.606 0.578 0.572 0.573 0.578 0.566 0.573 Degree of inequality aversion = 2.0 1969 1969-71 1969-73 1969-75 1969-77 1969-79 1969-81 1.109 0.949 0.889 0.951 0.974 1.035 1.100 1969 1969-71 1969-73 1969-75 1969-71 1969-79 1969-81 0.430 0.414 0.408 0.414 0.426 0.434 0.455 1969 1969-71 1969-73 1969-75 1969-77 1969-79 1969-81 0.375 0.364 0.360 0.363 0.373 0.381 0.396 1969 1969-71 1969-13 1969-75 1969-77 1969-79 1969-81 0.367 0.355 0.351 0.354 0.366 0.371 0.386 0.014 0.016 0.018 0.020 0.025 0.030 0.034 1.095 0.933 0.871 0.931 0.950 1.005 1.067 Degree of inequality aversion = 1.0 0.013 0.014 0.016 0.018 0.021 0.025 0.029 0.417 0.400 0.392 0.396 0.404 0.409 0.425 Degree of inequality aversion = 0.5 0.012 0.014 0.015 0.017 0.020 0.024 0.027 0.363 0.350 0.345 0.347 0.353 0.357 0.369 Degree of inequality aversion = 0.0 0.012 0.013 0.014 0.016 0.019 0.023 0.026 0.355 0.342 0.337 0.339 0.341 0.349 0.361 the between-group component increases to 15-25% of overall inequality for all unadjusted incomes [see Zandvakili (1987)]. This is partly due to a larger proportion of two income earners being among the male-headed families. The corresponding stability measures are presented in table 3. Again, seven of the thirteen possible values are reported without any qualitative loss. 0 < R,,, < 1 in all cases. The following may be concluded from table 3: (i) There is a tendency for the profiles to fall and then level off as the number of aggregated years increases from one to thirteen. 128 E. Maasoumi, and S. Zandvukili, Generalized entropy measures of mobility Table 3 1969-1981 per capita family income (MIW): Overall Between Degree of inequality 1969 1969-71 1969-73 1969-75 1969-71 1969-79 1969-81 1.000 0.916 0.885 0.828 0.786 0.744 0.692 1.000 0.928 0.903 0.877 0.855 0.832 0.813 1969 1969-71 1969-73 1969-75 1969-77 1969-79 1969-81 - 1.000 0.932 0.911 0.885 0.867 0.855 0.840 - 1.000 0.927 0.904 0.879 0.864 0.852 0.839 aversion 0.033 0.034 0.037 0.039 0.045 0.052 0.056 1.000 0.907 0.873 0.822 0.772 0.723 0.670 1.000 0.918 0.887 0.823 0.786 0.747 0.700 - 1.000 0.912 0.881 0.851 0.824 0.801 0.779 - 1.000 0.949 0.930 0.907 0.887 0.860 0.841 = 0.5 0.967 0.896 0.873 0.844 0.820 0.801 0.782 aversion Women = 1.0 0.970 0.896 0.869 0.839 0.812 0.783 0.761 aversion Men = 2.0 0.987 0.900 0.868 0.811 0.167 0.723 0.671 0.033 0.035 0.038 0.041 0.047 0.054 0.058 Degree of inequality 1969 1969-71 1969-73 1969-75 1969-77 1969-79 1969-81 aversion 0.030 0.032 0.035 0.038 0.043 0.049 0.052 Degree of inequality 1969 1969-71 1969-73 1969-75 1969-77 1969-79 1969-81 Within 0.013 0.016 0.017 0.018 0.020 0.021 0.021 Degree of inequality Income stability. 1.000 0.917 0.891 0.862 0.841 0.830 0.813 - 1.000 0.955 0.942 0.920 0.904 0.885 0.870 = 0.0 0.967 0.892 0.868 0.840 0.819 0.800 0.783 1.000 0.915 0.889 0.861 0.846 0.835 0.822 1.000 0.950 0.933 0.911 0.895 0.871 0.855 (ii) The profiles for households headed by men fall faster and further than those of women-headed households. (iii) These patterns are robust with respect to the choice of aggregation function, family size adjustment, and inequality measure. The fact that the profiles are becoming flatter is an indication that, although there have been some transitory movements in the size distribution of income, there is a lack of any permanent equalization. Further, while some equaliza- E. Muusoumi, und S. Zmdvakili, Generalized entropy measures of mobility tion has taken place within each group of households, inequality between and women-headed households has increased in absolute value. 129 men- 4. Mobility and income level Maasoumi and Zandvakili (1988) give inequality and mobility decompositions by age, education, and race. Similar decompositions by income level can reveal the aggregate impact of all such nonincome characteristics (including gender). It is anticipated that if the major causes of variation in incomes are transitory in nature, the length of time spent in any income class will be short. ‘Permanent’ income inequality changes will be very revealing in this context. The total sample is divided into seven income groups (GI-G7). The assignment to groups is on a one-time basis and according to the simple arithmetic mean income of the individual household over the thirteen-year period. These real income levels begin with mean incomes of less than $4,999, and increasing in increments of $5,000. The last group contains mean incomes of $35,000 or more. Short-run inequalities and their decompositions based on income level are given in table 4. There are several recognizable patterns. The ‘between-group’ inequality has increased steadily over this period. The ‘within-group’ component of inequality fluctuates around a relatively constant mean value. The observed patterns suggest that the nonincome differences do contribute to the increase in between-group inequality. Over 70% of women-headed households earn less than $15,000. Of course, this is confounded by the differential impact of inflation on different income groups (we use real incomes). Overall long-run inequality levels have risen after an initial decline. Its decomposition by income level shows that, as Z,(S) has changed, its betweengroup component has increased uniformly. At the same time the within group inequality has decreased steadily. This change has been dramatic so that in the later years the between-group component is larger than the within component. These changes include the well-known life cycle and human capital effects and are not inconsistent with the cummulative effects predicted by discrimination theories. The long-run within-group inequalities reveal a falling trend for each of the seven income groups. This is anticipated since transitory components are smoothed out and individual incomes have approached group mean incomes in the long run. These long-run grouping observations are somewhat sensitive to the family size. Within-group aggregate income inequalities are noticably smaller when income is not adjusted for family size, and there is generally less inequality within the higher income groups. The stability profiles in table 6 reveal much higher degrees of permanent equalization within income groups than was observed for the gender groups of the last section. Note that as the stability profiles of the whole sample flatten, --_. 0.367 0.402 0.395 0.448 0.492 0.483 0.549 1969 1971 1973 1975 1977 1979 1981 ^. 0.375 0.407 0.404 0.456 0.494 0.505 0.571 1969 1971 1973 1975 1977 1979 1981 .,I_. 0.430 0.466 0.464 0.531 0.578 0.613 0.706 1969 1971 1973 1975 1977 1979 1981 ._ - 1.109 1.049 1.002 1.636 1.569 1.895 2.441 1969 1971 1973 1975 1977 1979 1981 Overall - _ ,_, 0.272 0.264 0.224 0.231 0.219 0.212 0.249 0.103 0.144 0.180 0.225 0.275 0.293 0.322 _ 0.270 0.268 0.230 0.242 0.238 0.218 0.256 0.316 0.306 0.259 0.272 0.262 0.267 0.322 0.114 0.161 0.205 0.259 0.316 0.346 0.384 0.0% 0.134 0.165 0.207 0.254 0.266 0.293 0.958 0.824 0.692 1.225 1.045 1.271 1.716 Within 0.151 0.224 0.310 0.412 0.524 0.624 0.725 Between 1969-81 Table 4 - G2 1.429 0.560 0.415 0.539 0.487 0.683 0.985 0.413 0.334 0.284 0.324 0.310 0.374 0.490 0.356 0.299 0.263 0.292 0.282 0.331 0.432 0.402 0.398 0.348 0.333 0.343 0.345 0.382 . - 0.339 0.290 0.261 0.284 0.274 0.320 0.424 .---- 0.391 0.323 0.272 0.290 0.306 0.311 0.362 0.389 0.325 0.279 0.291 0.303 0.310 0.361 Degree of inequality aversion = 0.0 0.396 0.388 0.336 0.326 0.337 0.332 0.364 Degree of inequality aversion = 0.5 0.429 0.413 0.352 0.354 0.364 0.353 0.393 0.421 0.356 0.302 0.312 0.322 0.335 0.394 0.668 0.626 0.430 0.459 0.479 0.544 0.679 Degree of inequality aversion = 1.0 0.879 0.672 0.524 0.950 0.633 0.639 0.722 G3 Short-run inequalities. Degree of inequality aversion = 2.0 Gl per capita family income: --_-,/_ 0.349 0.367 0.313 0.293 0.263 0.271 0.311 0.353 0.373 0.314 0.294 0.260 0.269 0.315 0.385 0.411 0.336 0.314 0.273 0.286 0.342 0.608 0.725 0.479 0.449 0.365 0.460 0.525 G4 0.243 0.235 0.192 0.237 0.214 0.196 0.236 0.233 0.228 0.188 0.224 0.205 0.191 0.234 0.236 0.231 0.190 0.224 0.208 0.194 0.252 0.282 0.281 0.217 0.274 0.290 0.227 0.570 G5 0.263 0.246 0.214 0.215 0.176 0.200 0.222 _ 0.243 0.230 0.202 0.205 0.167 0.192 0.218 0.240 0.227 0.199 0.204 0.165 0.193 0.227 0.293 0.265 0.220 0.236 0.178 0.232 0.310 G6 0.187 0.228 0.202 0.223 0.245 0.191 0.233 0.179 0.219 0.192 0.212 0.224 0.183 0.221 0.179 0.224 0.190 0.211 0.218 0.183 0.225 0.209 0.450 0.212 0.242 0.242 0.211 0.299 G7 1.109 0.949 0.889 0.951 0.974 1.035 1.100 0.430 0.414 0.408 0.414 0.426 0.434 0.455 0.375 0.364 0.360 0.363 0.373 0.381 0.396 0.367 0.355 0.351 0.354 0.366 0.371 0.386 1969 1969-71 1969-73 1969-75 1969-77 1969-79 1969-81 1969 1969-71 1969-73 1969-75 1969-77 1969-79 1969-81 1969 1969-71 1969-73 1969-75 1969-77 1969-79 1969-81 1969 1969-71 1969-73 1969-75 1969-11 1969-19 1969-81 Overall 0.096 0.117 0.137 0.159 0.186 0.210 0.235 0.103 0.128 0.150 0.175 0.204 0.231 0.257 0.114 0.144 0.170 0.200 0.233 0.264 0.294 0.151 0.208 0.255 0.311 0.365 0.425 0.481 Between 1969-81 0.270 0.238 0.214 0.195 0.180 0.161 0.152 0.272 0.236 0.210 0.188 0.169 0.150 0.139 0.316 0.270 0.238 0.213 0.192 0.171 0.161 0.958 0.741 0.634 0.640 0.609 0.610 0.619 Within G2 (MIW): 1.429 0.708 0.504 0.404 0.341 0.313 0.308 0.413 0.311 0.261 0.225 0.200 0.181 0.179 0.356 0.274 0.236 0.205 0.182 0.164 0.162 0.402 0.339 0.298 0.263 0.246 0.230 0.220 0.339 0.261 0.226 0.197 0.174 0.153 0.153 Degree of inequality aversion 0.396 0.334 0.294 0.260 0.245 0.229 0.219 Degree of inequality aversion 0.429 0.354 0.309 0.275 0.259 0.241 0.232 = = 0.0 0.5 Degree of inequality aversion = 1.0 0.879 0.610 0.486 0.490 0.444 0.422 0.401 inequality. 0.391 0.318 0.256 0.219 0.191 0.171 0.165 0.389 0.320 0.262 0.225 0.196 0.172 0.163 0.421 0.346 0.282 0.242 0.210 0.180 0.167 0.668 0.548 0.425 0.348 0.303 0.270 0.244 G3 Long-run Degree of inequality aversion = 2.0 Cl per capita family income Table 5 0.349 0.335 0.296 0.262 0.214 0.182 0.173 0.353 0.342 0.304 0.269 0.219 0.185 0.175 0.385 0.374 0.328 0.290 0.233 0.192 0.180 0.608 0.601 0.506 0.426 0.322 0.274 0.247 G4 0.243 0.207 0.188 0.176 0.168 0.144 0.136 0.233 0.199 0.182 0.168 0.157 0.135 0.128 0.236 0.198 0.181 0.166 0.150 0.129 0.122 0.282 0.230 0.208 0.188 0.166 0.142 0.147 G5 0.263 0.223 0.212 0.190 0.169 0.147 0.135 0.243 0.208 0.198 0.179 0.161 0.139 0.127 0.240 0.204 0.193 0.176 0.158 0.135 0.122 0.293 0.240 0.222 0.208 0.188 0.160 0.146 G6 0.187 0.187 0.179 0.173 0.173 0.161 0.152 0.179 0.177 0.169 0.163 0.160 0.150 0.138 0.179 0.176 0.166 0.159 0.154 0.143 0.130 0.209 0.261 0.222 0.200 0.186 0.172 0.155 G7 132 E. Maasoumi, and S. Zanduakili. Generalized entropy treasures of mobiliiy ? _ II E. Muusoumi. and S. Zunduukili, Genercrlized entropy measures of mobilit,: 133 the corresponding within-group profiles continue to fall. In our view some equalization has occurred, but this is mostly confined to within income groups. 5. Conclusion A family of stability profiles was introduced and decomposed by gender and income level. The sensitivity of such profiles with respect to the choice of income aggregation function, the weights given to annual incomes in such aggregation, measure of inequality, and the length of time was investigated within certain limits. It was learned that the stability profiles generally flatten after a decline in the earlier years. The decomposition of both incomes and the stability values by gender revealed greater degree of equalization among households headed by men. Income level decompositions show income movements within each group and not across them. This is partly because we have studied a panel grouped by incomes and partly a reflection of nonincome effects. Inference about the U.S. on the basis of the Michigan data must be made with caution. References Atkinson, A.B., 1970, On the measurement of inequality, Journal of Economic Theory 2, 244-263. Bourguignon, F., 1979. Decomposable income inequality measures, Econometrica 47, 901-920. Creedy, J.. 19X5, Dynamics of income distribution (Basil Blackwell, London). Geweke. J.. R.C. Marshall. and G.A. Zarkin, 1986, Mobility indices in continuous time Markov chains, Econometrica 54, 1407-1423. Hart. P.E., 1976. The dynamics of earnings. 1963-1973, Economic Journal 86, 541-565. Kakwani, N., 1984, The relative deprivation curve and its applications, Journal of Business and Economic Statistics 2, 384-405. Lillard, L. and R. Willis, 1978. Dynamic aspects of earnings mobility, Econometrica 46, 985-1012. Maasoumi. E., 1986. The measurement and decomposition of multi-dimensional inequality. Econometrica 54, 991-997. Maasoumi, E., 1989. Composite indices of income and other developmental indicators: A general approach, Research on Income Inequality, forthcoming. Maasoumi, E. and G. Nickelsburg, 1988, Multivariate measures of well being and an analysis of inequality in the Michigan data. Journal of Business and Economic Stati&cs 6, 327-334. Maasoumi. E. and S. Zandvakili. 1986. A class of generalized measures of mobilitv with applications, Economics Letters 22, 97-102. Maasoumi, E. and S. Zandvakili, 1988, Mobility profiles and time aggregates of individual incomes, Research on Income Inequality, forthcoming. Prais, J.. 1955. Measuring social mobility. Journal of the Royal Statistical Society A 118, 55-66. Ram, R.. 19X2. Composite indices of physical quality of life, basic needs fulfillment, and income: A ‘principal component’ representation, Journal of Development Economics 11, 227-247. Rosen, S.. 1984, Comment on Kakwani (1984). Shorrocks. A.F., 1976, Income mobility and the Markov assumption, Economic Journal 86, 566-577. Shorrocks. A.F.. 1978, Income inequality and income mobility, Journal of Economic Theory 19. 376-393. Shorrocks, A.F., 1981, Income stability in the United States, Ch. 9 in: N.A. Klevmarken and J.A. Lybeck. eds.. The statics and dynamics of income (Tieto Ltd, Bristol). Shorrocks. A.F., 1984. Inequality decomposition by population subgroups, Econometrica 52. 1369-1385. Theil, H., 1967. Economics and information theory (North-Holland, Amsterdam). U.S. Bureau of the Census, Current population reports: Consumer income, P-60 (various years). Zandvakili, S.. 1987. The measurement of mobility and income stability: An information theoretic approach, Ph.D. thesis (Indiana University, Bloomington, IN).