Largest-remainder apportionment is used rather than rounding each n_h independently, because rounding can miss the target total by several units; ties break on the lowest stratum index so the two language arms agree exactly.
References
Cochran (1977), Sampling Techniques, 3rd edition, Section 5.3, Corollary 2, "stratification with proportional allocation of the n_h", n_h/n = N_h/N. Chapter 5 read from the scanned original. Cross-checked against the reference implementation in the CRAN package samplingbook 1.2.4, whose stratasamp(type = "prop") sets wh <- Nh/N.
Examples
Propalloc(Nh = c(1, 2, 3, 4, 5, 6, 7, 8), n = 5L)
#> $nh
#> [1] 0 0 0 1 1 1 1 1
#>
#> $nh_exact
#> [1] 0.1388889 0.2777778 0.4166667 0.5555556 0.6944444 0.8333333 0.9722222
#> [8] 1.1111111
#>
#> $Wh
#> [1] 0.02777778 0.05555556 0.08333333 0.11111111 0.13888889 0.16666667 0.19444444
#> [8] 0.22222222
#>
#> $fraction
#> [1] 0.1388889
#>
#> $N
#> [1] 36
#>
#> $n
#> [1] 5
#>
#> $L
#> [1] 8
#>
#> $method
#> [1] "Proportional allocation (largest remainder)"
#>
#> attr(,"class")
#> [1] "morie_rich_result" "list"
