¹ Hauser, P. M., ‘Aging and world-wide population change’, in Binstock, R. H. and Shanas, E. (eds), Handbook of Aging and the Social Sciences, Van Nostrand, New York (1976), pp. 59–86.Google ScholarGrundy, E., ‘Demography and old age’, Journal of the American Geriatrics Society, 31 (1983), 325–332.CrossRefGoogle ScholarPubMed

² Thompson, W. S., ‘Population’, American Journal of Sociology, 34 (1929), 959–975CrossRefGoogle Scholar, presented one of the earliest accounts of demographic transition theory. See also Beaver, S. H., Demographic Transition Theory Reinterpreted, Lexington Books, Lexington, Mass. (1975)Google Scholar, and Cutright, P. and Hargens, L., ‘The threshold hypothesis: evidence from less developed Latin American countries’, Demography, (1984), 459–473CrossRefGoogle ScholarPubMed, for more recent supportive arguments in favour of the demographic transition model.

³ Coale, A. J., ‘The demographic transition reconsidered’, International Population Conference, Vol. I, Liège, Belgium (1973), pp. 53–73Google Scholar; Teitelbaum, M. S., ‘Relevance of demographic transition for developing countries’, Science, 188 (1975), 420–425CrossRefGoogle ScholarPubMed; Laslett, P., ‘Societal development and aging’ in Binstock and Shanas (eds), op. cit., pp. 32–58.Google Scholar

⁴ Gilleard, C. J., Gurkan, A. A. and Gilleard, E., ‘Ageing in Turkey: patterns provisions and prospects’, in Butler, A. (ed.), Ageing: Recent Advances and Creative Responses, Croom Helm, London (1985).Google Scholar

⁵ Demeny, P. and Shorter, F. C., Estimating Turkish Mortality, Fertility and Age Structure, University of Michigan Population Study Center, Ann Arbor (1966).Google Scholar

⁶ The demographic indicators have been collated from State Institute of Statistics (SIS), 1980 Census of Population: Social and Economic Characteristics of Population, 1 Ankara (1983–1984), and Cerit, S., ‘Fertility estimates for Turkey according to the 1980 general census of population derived by direct computation and the Brass method’, The Turkish Journal of Population Studies, 6 (1984), 75–86.Google ScholarPubMed The provincial figures for value added are taken from SIS, Gross Domestic Product of Turkey (by Provinces): Sources and Methods, Ankara (1980).

⁷ Most of the studies on inter-provincial migration in Turkey indicate that it is the rural-to-urban migration which dominates the flows, despite the existence of substantial urban-to-urban population movements, see for example Munro, J., ‘Migration in Turkey’, Economic Development and Cultural Change, 22 (1973), 634–653.CrossRefGoogle Scholar Studies which are worth noting in this respect are: Gedik, A., A causal analysis of the destination choice of village to province center migrants in Turkey: 1965–1970, unpublished Ph.D. thesis, University of Washington (1977)Google Scholar; Yener, S., Inter-Provincial Migration and Migrant Characteristics: 1965–1970Google Scholar, State Planning Organisation, Ankara (1977) (Turkish); Tekeli, I. and Erder, L., Internal Migration as a Process of Adjustment in the Structure of Settlements, University of Hacettepe Press, Ankara (1978) (Turkish)Google Scholar; and Doh, R., ‘Inter-provincial migration in Turkey and its socioeconomic background: a correlation analysis’, The Turkish Journal of Population Studies, 6 (1984), 49–61.Google ScholarPubMed

⁸ United Nations, The Aging of Population and Its Economic and Social Implications, Population Studies, No. 26; New York (1956).

⁹ SIS, Statistical Yearbook of Turkey: 1983, Ankara, (1983), p. 43.

¹⁰ Brass, discussing the use of inter-censal survivorship as a means of calculating mortality rates and their use to develop life tables for a given population, states ‘the most serious deviations are due to fluctuations in the completeness of census taking and migration: … age errors are less important since they do not cause a systematic bias’ [Brass, W., ‘Indirect methods of estimating mortality illustrated by applications to Middle East and North African data’, in UN Economic Commission for Western Africa, The Population Framework: Data Collection, Demographic Analysis, Population and Development, Population Division, ECWA, Beirut (1978), pp. 121–165].

¹¹ The demographic variables used in this study were calculated using data from the 1980 census [SIS (1983–1984), op. cit.]. These variables and the labels by which they have been referred to throughout the study, are as follows.

URBAN – proportion of the population of settlements with more than 50,000 inhabitants, PROMAN – proportion of the economically active population engaged in manual work in the manufacturing industries, PROVIL – proportion of population living in villages, SURVIV – measure of late life mortality. PRO65 – proportion of population aged 65 +, IMR – infant mortality rate.

PROMAN has been used as a proxy for industrial development, since it appears to be a variable closely associated with other measures of industrial development [see, for example, Hacihasanoglu, B., Rankings and a Development Index for Provinces, SPO, Ankara (1980)Google Scholar (Turkish)]. Of the remaining demographic indicators, the total fertility rate (TFR) has been taken from Cerit, op. cit., and the measure for inter-provincial net migration (NETMIG) has been calculated using data from SIS, 1980 Census of Population: Domestic Migration by Permanent Residence, Ankara (1985).

The two variables representing household resources (proportion of dwellings with internal toilet facilities (PROWC) and electricity supply PROELEC)) have been taken from the 1975 census [SIS, Census of Population: Social and Economic Characteristics of Population, Ankara (1978–9)], while those representing health resources (number of hospital beds (HOSBED) and health personnel (HELPER) per 100,000 people) from the 1983 Yearbook [SIS, op. cit.]. The two rural development indices of intra-village specialisation of labour (VILSPE) and technical under-development index (AGDEVIND) were taken from Ministry of Village Affairs and Co-operatives, Village Inventory Studies, Ankara (1984), and Gurkan, A. A. ‘The regional structure of agricultural production in Turkey: a multivariate perspective’, METU Studies in Development, forthcoming (1985), respectively. The temperature variability (TVAR) representing the extent of temperature change during the year has been calculated by expressing the difference between the yearly minimum and maximum temperatures as a ratio of the average of the two figures. The data have been taken from General Directorate of Meteorology, Meteorology Yearbook: 1981, Ankara (1983), and are averages computed over the years of existence of the stations located in provincial centres.

¹² Of course current socio-economic indices can only account for the dispersion of the elderly across the provinces at a descriptive level since the historical development of the provinces will have contributed to the fertility and mortality rates of the various cohorts that currently make up their population structures. However, the relative importance of these historical forces will be reflected and will limit the explanatory variance that current indicators can presently account for. For a critical appraisal of cross-sectional approach to the analysis of demographic change see Simon, J. L., The Effects of Income on Fertility, Carolina Population Center, Monograph 12, University of North Carolina, NC (1974).Google Scholar

¹³ Spearman's rank-order correlation coefficients between the residuals obtained from regression and the independent variables were used for testing the assumption of homoscedasticity [see Johnston, J., Econometric Methods, McGraw-Hill, London, second edition (1972)]Google Scholar and the Kolmogorov-Smirnov Z-statistic was used for testing the assumption of normality of the distribution of the error terms. The multicollinearity diagnostics are those of Belsey et al. and Belsley [Belsley, D. A., Kuh, E. and Welsch, R. E., Regression Diagnostics, Identifying Influential Data and Sources of Collinearity, Wiley, New York (1980)CrossRefGoogle Scholar; Belsley, D. A., ‘Assessing the presence of harmful collinearity and other forms of weak data through a test for signal-to-noise’, Journal of Econometrics, 20 (1982), 211–253].CrossRefGoogle Scholar

¹⁴ When multicollinearity exists, the standard errors of the estimated coefficients are usually very large and the coefficients themselves are sensitive to small changes [Belsley, et al. , op. cit.].Google Scholar

¹⁵ A more detailed technical report describing the tests and the estimation techniques employed in the study has been deposited with the Editors of the Journal. A copy of the report could also be obtained from the following address: Dr A. A. Gurkan, Department of Economics, Middle East Technical University (ODTU), Ankara, Turkey.

¹⁶ Ridge and principal components regression, among others [see Driscoll, M. F. and Reynolds, D. A., Biased Regression, Technical Report No. ACS–4, Academic Computing Services, Arizona State University, Arizona (1979)]Google Scholar, have been suggested as possible solutions that could mitigate the harmful effects of multicollinearity. Because of the ad hoc nature of the techniques, some find their use problematic [see, for example, Pagel, M. D. and Lunneborg, C. E., ‘Empirical evaluation of ridge regression’, Psychological Bulletin, 97 (1985), 342–355CrossRefGoogle Scholar; Fomby, T. B. and Carter-Hill, R., ‘Deletion criteria for principal components regression analysis’, American Journal of Agricultural Economics, 60 (1978), 524–527].CrossRefGoogle Scholar However, others have demonstrated the usefulness, especially, of principal components regression in applied research settings [see Massy, W. F., ‘Principal components regression in exploratory statistical research’, Journal of the American Statistical Association, 60 (1965), 234–256CrossRefGoogle Scholar; Greenberg, E., ‘Minimum variance properties of principal component regression’, Journal of the American Statistical Association, 70 (1975) 194–197CrossRefGoogle Scholar; and Mittlehammer, R. C., Young, D. L., Tasanasanta, D. and Donnely, J. T., ‘Mitigating the effects of multicollinearity using exact and stochastic restrictions: the case of aggregate agricultural production function in Thailand’, American Journal of Agricultural Economics, 62 (1980), 199–210].CrossRefGoogle Scholar The technique essentially involves deleting the principal components, extracted from the full set of independent variables, with relatively smaller eigenvalues from the regression analysis and carrying out the estimation with only those that remain; the estimates of the coefficients of the original variables are then derived using their loadings on the components.

¹⁷ It has been shown that principal components regression can be cast into a restricted least squares problem [Fomby and Carter-Hill, op. cit.], which then allows the use of mean square error tests to determine the number of components to be deleted from analysis. These tests have been developed by Toro-Vizcarronda and Wallace and Wallace [Toro-Vizcarronda, C. E. and Wallace, T. D. ‘A test of the mean square error criterion for restrictions in linear regression’, Journal of the American Statistical Association, 63 (1968), 558–572Google Scholar; Wallace, T. D., ‘Weaker criteria and tests for restrictions on linear regression’, Econometrica, 49 (1972), 689–698]CrossRefGoogle Scholar, and the tables of the non-central F distribution to be used in the tests can be found in Wallace, T. D. and Toro-Vizcarronda, C. E., ‘Tables for the mean square error test for exact linear restrictions in regression’, Journal of the American Statistical Association, 64 (1969), 1649–1663.CrossRefGoogle Scholar

¹⁸ The sources for these variables have been mentioned in note 11.

¹⁹ The links between international differences in proportions of elderly and economic development have been drawn by Hauser, , op. cit.Google Scholar, Cowgill, D. O. and Holmes, L. D., Ageing and Modernization, Appleton Century, New York (1972)Google Scholar; Hendricks, J. and Hendricks, C. D., Aging in Mass Society, Winthrop, Cambridge, Mass. (1977)Google Scholar; between late life mortality and economic development by Preston, S. H., Mortality Patterns in National Populations, Academic Press, New York (1976).Google Scholar One of the most recent and thorough analyses, conducted by Weatherby et al., concludes ‘countries with higher per capita income levels tend to have lower age specific death rates in the age-intervals 65–69 through 80–84, than do countries with lower income levels’ [Weatherby, N. L., Nam, C. B. and Isaac, L. W., ‘Development, inequality, health care and mortality at the older ages; a cross national analysis’, Demography, 20 (1983), 27–41].CrossRefGoogle Scholar

²⁰ Shorter, F. C. and Macura, M., Population Increase in Turkey 1935–1975.: Trends in Fertility and Mortality, Yurt Yayinlari, Ankara, second edition (1983)Google Scholar (Turkish).

²¹ Kagitcibasi, C., ‘Old age security value of children: cross-national socio-economic evidence’, Journal of Cross-Cultural Psychology, 13 (1982), 29–42.CrossRefGoogle Scholar

²² Stolnitz, G. J., ‘Three to five challenges to demographic research’, Demography, 20 (1983) 424–425.CrossRefGoogle Scholar

²³ Beaver, , op. cit.Google Scholar