Relational model life tables (cont.)

If any two lx columns are logit-transformed and then plotted together they do form a tolerably straight line. Let's take an example:

Age x CD West 30   UN LA 60    
  100000 lx propn. Yx
logit lx
100000 lx propn. Yx
logit lx
5 61196 0.6120 -0.2278 85881 0.8588 -0.9027
10 58141 0.5814 -0.1643 84901 0.8490 -0.8634
15 55864 0.5586 -0.1178 84480 0.8448 -0.8472
20 53002 0.5300 -0.0601 83900 0.8390 -0.8254
25 49614 0.4961 0.0077 83204 0.8320 -0.8001
30 46072 0.4607 0.0787 82423 0.8242 -0.7727
35 42361 0.4236 0.1540 81472 0.8147 -0.7405
40 38624 0.3862 0.2316 80293 0.8029 -0.7024
45 34968 0.3497 0.3102 78743 0.7874 -0.6548
50 31396 0.3140 0.3908 76609 0.7661 -0.5932
55 27280 0.2728 0.4902 73294 0.7329 -0.5048
60 22764 0.2276 0.6109 68218 0.6822 -0.3819
65 17384 0.1738 0.7793 60627 0.6063 -0.2158
70 12025 0.1202 0.9951 50383 0.5038 -0.0077
75 6931 0.0693 1.2987 37619 0.3762 0.2529
80 3073 0.0307 1.7256 23960 0.2396 0.5774
85 937 0.0094 2.3302 12350 0.1235 0.9799

The table above shows the lx column of two different mortality patterns – the CD West pattern at e0=30yrs and the UN Latin American pattern at e0=60yrs.

The columns have been transformed firstly to proportions and then to logits.  Note how the logit value changes from negative to positive as the lx proportion drops below 0.5.

Click to reveal a relational plot of these two patterns

 

Relational Model

A trendline has been fitted. The "tolerably straight" line is apparent although it is clear that there is a distinct lack of linearity on the extreme left of the graph. The reason for this departure from linearity is that we have purposely chosen two very different mortality patterns to compare and so, in effect, we are pushing the linear relationship to its limits - but even so the relationship is quite good.