Air pressure as a function of elevation, p. 4. Residuals from
    the quadratic regression of pressure in mmHg on elevation.

    ------
          [1,]        [2,]        [3,]        [4,]        [5,]        [6,] 
    40.2976742  36.8360934  33.3025758  29.5971213  26.2197300  22.6704017 
          [7,]        [8,]        [9,]       [10,]       [11,]       [12,] 
    19.3491366  15.8559346  12.3907957   9.0537200   5.7447073   2.5637578 
         [13,]       [14,]       [15,]       [16,]       [17,]       [18,] 
    -0.2891286  -3.3139519  -6.0107120  -8.4794091 -11.0200430 -13.3326138 
         [19,]       [20,]       [21,]       [22,]       [23,]       [24,] 
   -15.8171215 -17.7735660 -19.8019474 -23.4745209 -26.8348419 -29.9829105 
         [25,]       [26,]       [27,]       [28,]       [29,]       [30,] 
   -32.6187265 -34.8422901 -40.9763206 -40.1040390 -34.0254453 -24.3405395 
        [31,]       [32,]       [33,]       [34,]       [35,]       [36,] 
   -12.3493216   1.1482085  15.4520506  29.0622048  40.4786712  48.6014497 
         [37,]       [38,]       [39,]       [40,] 
    52.1659430  34.7556847  -6.5293252 -73.6290865 

    - - - - - - -

    The residuals are not as large as for the simple linear
    regression, and the pattern is not quite the same,
    but there is still a pattern: the residuals are steadily
    decreasing until the 27th, then they increase up to the
    37th, then they decrease. So there are just two turning
    points, far fewer than there should be for i.i.d.
    residuals. One could consider higher order polynomials
    in the elevation, but a polynomial regression model won't
    be satisfactory for all values of elevation h, such as
    very large values, so on the next page we'll consider
    a different non-polynomial regression.