fquantile bug

[zander-prinsloo]

Hi Sebastian,

I picked up an issue with fquantile and compare it to Hmisc::wtd.quantile() in the example below. I realize that the output from these two functions should not be exactly the same because they are not using the same approach to calculating the quantiles, so the small differences from most of the quantiles are not what I am referring to.

The issue I am pointing at is the 60% quantile from collapse::fquantile(), which decreases and then jumps back up for 70%. For the example below, this issue persists even when using different types (i.e. type = 5, etc.). Interestingly, the final example below where fquantile( x = c(1:100), w = rep(1, 100), probs = seq(from = 0.1, to = 1, by = 0.1)) the issue seems to disappear.

Thank you very much! As always, I am really enjoying the collapse package and appreciate all your work on it greatly.

Zander

library(collapse, Hmisc)

x <- c(0.3031276, 0.4571868, 0.5611382, 0.6589519, 0.7138266, 0.7599011,
             0.7859537, 0.8227337, 0.8861251, 0.9227733, 0.9741309, 1.0043999,
             1.0527802, 1.1004243, 1.1344353, 1.1543153, 1.1986660, 1.2225957,
             1.2467686, 1.2881969, 1.3103023, 1.3470067, 1.3717622, 1.4171236,
             1.4481890, 1.4767816, 1.5113053, 1.5520433, 1.5766412, 1.6206641,
             1.6614114, 1.6988976, 1.7187559, 1.7445536, 1.7655008, 1.7951064,
             1.8154464, 1.8412325, 1.8843781, 1.9022643, 1.9349266, 1.9775958,
             2.0209325, 2.0538855, 2.0880485, 2.1341179, 2.1662982, 2.2073586,
             2.2297966, 2.2696783, 2.3054806, 2.3522053, 2.4107709, 2.4608277,
             2.4788487, 2.5041681, 2.5652320, 2.6004631, 2.6632136, 2.7169775,
             2.7553121, 2.8365738, 2.9027606, 2.9793340, 3.0432024, 3.1047007,
             3.1667340, 3.2568974, 3.3373767, 3.4326348, 3.5228778, 3.6003411,
             3.6967467, 3.8003151, 3.9020071, 3.9976514, 4.1647284, 4.2741119,
             4.3598892, 4.4362847, 4.5396775, 4.6069366, 4.7372277, 4.9295803,
             5.0362082, 5.1420157, 5.2902402, 5.5672972, 5.7774503, 6.2099230,
             6.7065727, 6.9770491, 7.2619571, 7.8597135, 8.3030427, 8.8661065,
             9.9074310, 10.7280315, 12.6459594, 18.0195166)

w <- c(148.832031, 861.869185, 753.803590, 787.644447, 680.662491,
            889.665800, 835.005236, 744.138014, 631.031320, 59.672307,
            211.059592, 180.225505, 481.189884, 428.716380, 963.777711,
            768.157068, 620.819895, 55.915707, 744.457704, 80.550037,
            596.757105, 293.466754, 845.799925, 127.239379, 65.016717,
            982.369334, 538.933227, 304.913201, 1000.549589, 229.999551,
            900.345681, 715.797532, 926.045942, 445.912266, 57.284686,
            200.823635, 311.681956, 288.483657, 715.706993, 181.051739,
            988.774738, 558.520419, 146.126589, 706.082199, 443.189061,
            194.593867, 623.515651, 281.936126, 4.341623, 606.789155,
            929.317259, 220.582461, 356.963444, 69.051047, 267.023953,
            870.515426, 145.374869, 285.938089, 343.880965, 227.054974,
            874.782554, 899.026458, 551.543717, 838.777356, 28.366997,
            668.319839, 791.231414, 616.596092, 211.014267, 754.605834,
            956.967726, 566.716436, 935.859162, 64.795456, 619.012640,
            205.367614, 979.031488, 942.429357, 65.017801, 222.496672,
            465.333957, 305.959686, 595.160696, 361.537173, 359.102206,
            842.988743, 649.262715, 385.603889, 416.167539, 512.941754,
            911.603964, 424.459892, 823.561780, 856.990370, 534.810471,
            870.347438, 392.741960, 965.569166, 663.887435, 338.334480)

collapse::fquantile(
  x = x, 
  w = w, 
  probs = seq(from = 0.1, to = 1, by = 0.1)
)
#>        10%        20%        30%        40%        50%        60%        70% 
#>  0.8381894  1.0836237  1.6189591  1.9205753  2.3487439  1.3716575  3.6852311 
#>        80%        90%       100% 
#>  4.7764544  7.3361690 18.0195166
Hmisc::wtd.quantile(
  x = x, 
  w = w, 
  probs = seq(from = 0.1, to = 1, by = 0.1)
)
#>        10%        20%        30%        40%        50%        60%        70% 
#>  0.8227337  1.2467686  1.6206641  1.9349266  2.3522053  2.9793340  3.6003411 
#>        80%        90%       100% 
#>  4.7372277  7.2619571 18.0195166
collapse::fquantile(
  x = c(1:100), 
  w = rep(1, 100), 
  probs = seq(from = 0.1, to = 1, by = 0.1)
)
#>   10%   20%   30%   40%   50%   60%   70%   80%   90%  100% 
#>  10.9  20.8  30.7  40.6  50.5  60.4  70.3  80.2  90.1 100.0

[SebKrantz]

Thanks, yeah this is strange. Looks like something is rounded off somewhere where it shouldn't. In general, the algorithm is a bit different from Hmisc, so the results are not identical by design, but this is of course a bug.

[SebKrantz]

This has been fixed now in 2.0.9. Thanks for flagging. Will send it to CRAN today.