Page 331 - Special Topic Session (STS) - Volume 3
P. 331

STS547 Maarten C. et al.
               total is Māori. For the Census the estimated number is 721,971. For the MOH
               it is 640,711.
               3.2 Four registers
                   We now add the other two registers, DIA and MOE, to the analysis. Now
               the maximal model is [ABCd][ABDc] [ACDb][BCDa] [ABcd][ACbd] [ADbc][BCad]
               [BDac][CDab] [Abcd][Bacd] [Cabd] [Dabc][abcd]. Notice that, as for the two
               registers, a capital variable label cannot be in the same interaction term as a
               lower case variable label, as these interactions cannot be estimated from the
               data. Notice that the assumptions become less and less demanding as more
               registers are involved. The number of unique individuals in the four linked
               registers is 4,401,282, and the estimated number missed by all registers is
               25,939, giving an estimated population size of 4,427,221.
                   The estimated numbers of Maori are displayed in Table 2. To arrive at a
               final  estimate  of  the  number  of  non-Māori  and  Māori  we  describe  two
               approaches,  both  using  the  concept  of  measurement  error.  Consider  the
               margins of the ethnicity variables a; b; c and d of the four registers. A statistical
               approach  to  measurement  error  is  to  make  use  of  a  latent  class  model
               (McCutcheon, 1987). See Table 3. In this latent class model, the first latent class
               is to be interpreted as the class for non-Māori, and the estimated probability
               of falling in this class is 0.826. The probability for the Māori class corresponds
               to an estimated Māori population size of about 770,000. Estimated conditional
               probabilities of being Māori for each latent class are also shown in Table 3;
               they are consistently low for the non-Māori latent class and high for the Māori
               latent class.
               Panel 1: Observed counts
                                                  C = 1                C = 0     Totals
                                       c = 0      c = 1      c = -     c = x
                 A = 1     a = 0     3,004,329    31,998    150,855    38,640   3,225,822
                           a = 1       108,192   435,468     12,402      4,377    530,439
                           a = -        16,512      2,769       894       435      20,160
                 A = 0     a = x       398,838   146,985     24,642         -     570465
                 Totals              3,527,871   617,220    188,793    43,452  4,377,336
               Panel 2: Fitted values under [Ac][ac][Ca]
                                           C = 1                C = 0           Totals
                                     c = 0       c = 1      c = 0     c = 1

                 A = 1    a = 0    3,170,298.4   33,791.2   38,619.1   411.6   3,243,120.3
                          a = 1     111,244.8  448,084.6     879.3    3,541.9   563,750.6
                 A = 0    a = 0     402,713.4   10,772.5    4,905.7    131.2    418,522.8
                          a = 1       14,131.1  142,848.3    111.7    1,129.2   158,220.3
                 Totals            3,698,387.7  635,496.6   44,515.8   5,213.9  4,383,613.7



                                                                  320 | I S I   W S C   2 0 1 9
   326   327   328   329   330   331   332   333   334   335   336