Plot_model predicts negative values for a proportional response variable

Question

I am new to the Sjplot plot_model function. I am trying to plot a linear model, where the response variable is a proportion (min 0, max 1), yet when plotting this model the predicted values are plotted for negative values too.

The simple model I use; sexpredictionadult <- lm(Propmort~Maletofemale,data=subset)

I try to correct this using +ylim, but this cuts part of the plot away. plot_model(sexpredictionadult,type="slope",terms=c("Maletofemale"))

plot_model(sexpredictionadult,type="slope",terms=c("Maletofemale"))+ylim(c(0,1))

structure(list(Maletofemale = c(0.749999997916667, 0.0123456790047249, 
0, NA, 0, 0, 0, 0.387499999515625, 0.3999999984, 0, 0.820512818408941, 
0.47712418269469, 0, 0.461538457988166, 0, 5.8e+08, 6.6e+08, 
0, 0, 0.19148936156632, 0.786885244611664, 0, 0.681415928902028, 
2.38e+09, 0.249999999107143, 0, 0, 0.35999999928, 5.3e+08, NA, 
0, 0, 0, 0.290540540344229, 0.34821428540338, NA, 0, 2.38e+09, 
0, 0, 0, 0.195402298625974, 3.66666660555556, 1.30769229763314, 
0, 0.363636361983471, 1.36842104903047, 0.312499998046875, 0, 
0.315789472022161, 0.499999997222222, 0.545454542975207, 0.619047616099773, 
0, 0.211864406600115, 0.866666660888889, 0.818181810743802, 7.3e+08, 
0.499999998076923, 0.102564102301118, 8e+08, 0, 0.999999991666667, 
0.405405405131483, 0, 1.06249999335938, 1.15789473074792, 0, 
0, NA, NA, NA, NA, 1.02777777635031, 0, 2.85714281632653, NA, 
0, 0, 0, NA, NA, 0, 0, 0, 0, 0.146245059230733, 0.52054794484894, 
0.938931296993182, 0.958333329340278, 0, 0.999999993333333, 1.73e+09, 
0.547619046315193, 0, 7.6e+08, NA, 0, NA, 0, 0, 0.365853657644259, 
0, 3.4e+08, 0, 0, 0, NA, NA, NA, 0.307692305325444, 0.34999999825, 
0.7368421033241, NA, NA, 0, 0, 0, 0, 0, 0, 0, 6e+08, 0, 0, 0, 
0, 0, 1.08333332881944, 0, 0, NA, 0.50561797695998, 0.428571425510204, 
0.666666661111111, 0, 0, 0, 0.77777777345679, NA, 0, 0, 0.8999999991, 
0.424242422956841, NA, 0.901639342784198, 0.653846151331361, 
4.8e+08, 0.99999999, 0.999999999363057, 0.821428568494898, 0.627272726702479, 
0.411764703460208, 0.349999999125, NA, 7.5e+08, 0.555555549382716, 
3.76923075956607, 0.232142856728316, 1.888888881893, 0.0487804876859012, 
0, 0, 0, 0, 0.478260869218652, 0, 0, 0.999999994444444, 0, 0.652542372328354, 
0, 0.655172412663496, 0.451612902983004, 0, 0, 0.636363630578513, 
0, 0.266666665777778, NA, 2.1e+08, 5.2e+08, 0, 0.714285709183674, 
1.15789473074792, 0, 0, NA, 0.404761903798186, 0, 1.06451612559834, 
0, NA, 0.07999999968, 0.303030302112029, 0.116279069497025, NA, 
NA, 0, 0, 0, NA, 0, NA, NA, 0, 0, 0, 0.622222220839506, 1.60439560263253, 
0, 0.631578945706371, 0, 0, 0.487179485930309, 0, 0, 5.4999999828125, 
1.94e+09, NA, NA, 0, 0, 0, NA, 3.09090906280992, 0.918367345064556, 
NA, NA, 5.56e+09, 4.8e+09, 1.49999998928571, 0, 4.8e+08, 0, 0.285714283673469, 
0, 0.256097560663296, 1.799999964, 0.562499996484375, 0.399999996, 
0.102040816118284, 0.71739130278828, 0.142011834277511, 0.514285712816327, 
0, 0.717948716107824), Propmort = c(0.6, 0.25, 1, 0.391304347826087, 
0.0689655172413793, 0.133333333333333, 0.928571428571429, 0.380952380952381, 
0.4375, 0.125, 0.0294117647058824, 0.258620689655172, 0, 0, 0.176470588235294, 
0.32, 0.0689655172413793, 0.857142857142857, 0.333333333333333, 
0.730769230769231, 0.521739130434783, 0.4, 0.461538461538462, 
0, 0.75, 0.0909090909090909, 1, 0, 0.1, 0.0476190476190476, 0.625, 
0.727272727272727, 0.375, 0.348484848484849, 0.26530612244898, 
0.2, 0.111111111111111, 0, 0, 0, 0.368421052631579, 0.0555555555555556, 
0, 0.181818181818182, 0, 0, 0, 0.666666666666667, 0.546511627906977, 
0.6, 0.666666666666667, 0.857142857142857, 0.466666666666667, 
0, 0.446808510638298, 0, 1, 0.384615384615385, 0, 0.142857142857143, 
0.0967741935483871, 0, 0.833333333333333, 0, 0.518518518518518, 
0, 0, 0.866666666666667, 1, 0.5, 0.2, 0.32258064516129, 0.588235294117647, 
0.209302325581395, 0.3, 0.666666666666667, 0, 0.740740740740741, 
0, 0, 0, 0.666666666666667, 0.166666666666667, 0.0384615384615385, 
0, 0, 0, 0, 0, 0, 0.205882352941176, 0, 0, 0.411764705882353, 
0, 0, 0, 0.0555555555555556, 0.535714285714286, 0.60655737704918, 
0.418181818181818, 0, 0, 0.454545454545455, 0.242990654205607, 
0.574074074074074, 0.75, 0, 0, 0, 0.666666666666667, 0.166666666666667, 
0.1875, 0.771084337349398, 0.780821917808219, 0.888888888888889, 
0, 0.32, 0.857142857142857, 0.516129032258065, 0.7, 0.277777777777778, 
0, 0.775, 0.357142857142857, 0.714285714285714, 0, 0, 1, 0.785185185185185, 
0, 0.75, 0.0326086956521739, 0.166666666666667, 0.428571428571429, 
0.777777777777778, 0.736842105263158, 0.846153846153846, 0, 0.615384615384615, 
0.75, 1, 0.315068493150685, 0.615384615384615, 0.333333333333333, 
0.0625, 0.666666666666667, 0, 1, 0.0344827586206897, 0.428571428571429, 
0.0444444444444444, 0, 0.4, 0.676470588235294, 0.0526315789473684, 
0, 0.136986301369863, 0, 0.0512820512820513, 0.5, 1, 0.805555555555556, 
0.6875, 0, 0, 0.0909090909090909, 0.458333333333333, 0.25, 0, 
0.539473684210526, 0, 0, 0.150943396226415, 0, 0.666666666666667, 
0, 0.5, 0.75, 0.386363636363636, 0, 0.0666666666666667, 0.0465116279069767, 
0, 0.181818181818182, 0, 0.0909090909090909, 0, 0, 0, 0.0526315789473684, 
0.1, 0.0833333333333333, 0.545454545454545, 0.764705882352941, 
0.133333333333333, 0.75, 0, 0.5, 0.9, 0.230769230769231, 0.454545454545455, 
0.397260273972603, 1, 0.722222222222222, 0.142857142857143, 0.117647058823529, 
0.181818181818182, 0.292682926829268, 0, 0.607142857142857, 0.307692307692308, 
0.363636363636364, 0.814814814814815, 0.444444444444444, 0.625, 
0, 0.0289855072463768, 0.483516483516484, 0.702702702702703, 
0.75, 0.19047619047619, 0, 0.181818181818182, 0.806451612903226, 
0, 0.75, 0, 0.260869565217391, 0, 0.0175438596491228, 0, 0, 0, 
1, 0.5, 0.173913043478261, 0.488372093023256, 0, 0.666666666666667, 
0, 0.666666666666667, 0, 0.411042944785276, 0.176470588235294, 
0.894736842105263, 0)), row.names = c(2L, 3L, 7L, 9L, 12L, 13L, 
16L, 21L, 22L, 23L, 24L, 25L, 28L, 29L, 30L, 31L, 32L, 37L, 39L, 
40L, 41L, 43L, 45L, 48L, 50L, 51L, 53L, 54L, 55L, 57L, 59L, 60L, 
61L, 63L, 67L, 70L, 71L, 73L, 74L, 75L, 77L, 78L, 79L, 80L, 81L, 
82L, 83L, 84L, 85L, 86L, 87L, 88L, 89L, 90L, 91L, 93L, 94L, 95L, 
96L, 98L, 100L, 103L, 104L, 108L, 109L, 110L, 111L, 113L, 114L, 
115L, 116L, 117L, 118L, 120L, 121L, 122L, 123L, 124L, 126L, 128L, 
129L, 131L, 133L, 140L, 144L, 145L, 148L, 149L, 150L, 151L, 152L, 
154L, 157L, 159L, 160L, 161L, 162L, 163L, 164L, 165L, 166L, 167L, 
168L, 170L, 171L, 173L, 174L, 177L, 178L, 179L, 180L, 182L, 184L, 
185L, 186L, 190L, 191L, 193L, 198L, 200L, 203L, 204L, 205L, 207L, 
209L, 210L, 212L, 213L, 214L, 215L, 220L, 221L, 222L, 223L, 224L, 
225L, 226L, 227L, 229L, 230L, 231L, 232L, 234L, 235L, 236L, 238L, 
239L, 240L, 241L, 244L, 245L, 246L, 247L, 249L, 251L, 252L, 253L, 
254L, 255L, 256L, 260L, 262L, 263L, 264L, 265L, 266L, 267L, 269L, 
271L, 272L, 277L, 278L, 280L, 281L, 282L, 283L, 284L, 285L, 291L, 
294L, 295L, 296L, 297L, 303L, 305L, 306L, 307L, 308L, 309L, 312L, 
316L, 317L, 319L, 320L, 321L, 322L, 324L, 325L, 329L, 330L, 331L, 
333L, 334L, 335L, 336L, 339L, 343L, 345L, 349L, 350L, 352L, 355L, 
356L, 357L, 359L, 361L, 362L, 364L, 366L, 367L, 368L, 369L, 370L, 
372L, 376L, 377L, 378L, 380L, 382L, 386L, 387L, 391L, 393L, 394L, 
395L, 396L, 397L, 399L, 400L, 403L, 404L, 407L, 409L, 410L, 412L, 
413L, 416L), class = "data.frame")

dataset:

It's hard to tell for sure without a reproducible example, but I think the problem would be with the model, and not `sjPlot`. Can you update the post so that we can fit a version of the model that creates this problem? — sjp, Aug 18 '22 at 19:49

score 0 · Answer 1 · answered Aug 18 '22 at 20:16

The issue here is not with sjPlot, but rather you are seeing a good example of why it is not a good idea to model proportion data using linear models. After fitting the model that you specified, I recreate your plot (using slightly different code, as type="slope" for me produced something different from what you posted). I also modify the plot to show the raw data:

# Fitting the model
sexpredictionadult <- lm(Propmort~Maletofemale,data=subset)

# Plotting the model with the data included
plot_model(sexpredictionadult,
           type="pred",
           terms=c("Maletofemale"),
           show.data = TRUE)

So we can see that the model is predicting values of the outcome variable rather poorly for higher values of maletofemale. Namely, the model thinks these values should be negative when in reality, we know they are bounded by zero. You should look into a model that more closely aligns with the nature of your data. Ben Bolker gives several options in the post I've linked below on the stats exchange, where the poster gives a data description that seems similar to the data you're working with:

https://stats.stackexchange.com/questions/560996/which-glmm-to-use-for-proportional-data

Thank you @sjp Since many of my responses have a zero value, perhaps I should look at zero-inflated beta regression? When looking for information on models with a proportional response variable, I either see articles about beta regressions and then about linear models with a binomial distribution, it is however unclear to me which one to choose in which situation — Beardedant, Aug 18 '22 at 20:53
It's unclear to me what you are modelling. If the proportions are made up of 0s and 1s, and you have access to that data, or you have the proportion scores and know how many observations each proportion comes from, then you can use logistic regression (GLM with a binomial family function). I haven't use beta regression much but it also seems like an option for you, although I hear having 0s and 1s can be a problem — sjp, Aug 18 '22 at 21:15

Plot_model predicts negative values for a proportional response variable

1 Answers1