Constructing and Analyzing a Nesting Dataset in R
Jairus Otana
2025-05-20
About the Dataset
For this activity, I worked with a dataset on White Browed Sparrow Weaver nesting, applying logistic regression and data visualization using R.
##Loading the libraries
library(dplyr)##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, unionlibrary(readxl)
library(corrplot)## corrplot 0.95 loadedlibrary(tidyr)
library(GGally)## Loading required package: ggplot2## Registered S3 method overwritten by 'GGally':
## method from
## +.gg ggplot2library(ggplot2)##Load Data
data <- read_excel("C:/Users/User/Downloads/Data rangu/Data rangu/data.xlsx")
View(data)
data## # A tibble: 915 × 8
## TS `P/A` TH CD CC BA DS NN
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 Mopane 1 10.6 9.4 70.2 8.94 0 14
## 2 Mopane 1 11.1 10.2 23.3 0.7 10 1
## 3 Mopane 0 10.6 9.95 29.2 1.53 10 0
## 4 Mopane 0 14 11.7 48.3 1.43 0 0
## 5 Mopane 0 10.2 8.55 41.4 1.88 10 0
## 6 Mopane 1 8.5 7.27 48.1 1.76 50 10
## 7 Mopane 1 10.5 8.99 61.4 1.47 10 4
## 8 Mopane 0 11.5 9.19 47.1 1.72 10 0
## 9 Mopane 0 13 11.6 40.8 1.22 10 0
## 10 Mopane 0 10.6 9.75 64.1 1.03 0 0
## # ℹ 905 more rowsSteps Taken
1. Data Loading & Cleaning
Loaded data using read_excel() and read.csv().
Converted the P/A (Presence/Absence) variable into a factor for binary classification.
data$`P/A` <- as.factor(data$`P/A`)- Logistic Regression Modeling Modeled nest presence using tree structural variables: Tree Height (TH), Canopy Depth (CD), Canopy Cover (CC), Basal Area (BA), and Damage Score (DS):
model <- glm(`P/A` ~ TH + CD + CC + BA + DS, data = data, family = binomial)
summary(model)##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = data)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.600539 0.331455 -4.829 1.37e-06 ***
## TH 0.151297 0.045991 3.290 0.001 **
## CD -0.039035 0.044741 -0.872 0.383
## CC 0.022358 0.005591 3.999 6.37e-05 ***
## BA 0.099349 0.085080 1.168 0.243
## DS -0.026155 0.004276 -6.117 9.55e-10 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1163.2 on 912 degrees of freedom
## Residual deviance: 773.8 on 907 degrees of freedom
## (2 observations deleted due to missingness)
## AIC: 785.8
##
## Number of Fisher Scoring iterations: 5##Checked model quality using AIC and interpreted odds ratios:
AIC(model)## [1] 785.7988exp(cbind(OR = coef(model), confint(model)))## Waiting for profiling to be done...## OR 2.5 % 97.5 %
## (Intercept) 0.2017877 0.1040817 0.3824559
## TH 1.1633423 1.0641990 1.2748632
## CD 0.9617168 0.8801620 1.0492515
## CC 1.0226101 1.0116467 1.0341130
## BA 1.1044513 0.9212414 1.3105013
## DS 0.9741841 0.9659280 0.9822820Interaction Effects
To examine whether combinations of tree features influence nesting more than individual ones, I added interaction terms:
interaction_model <- glm(`P/A` ~ TH * CD + BA * CC, data = data, family = binomial)
summary(interaction_model)##
## Call:
## glm(formula = `P/A` ~ TH * CD + BA * CC, family = binomial, data = data)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -4.730676 0.429975 -11.002 < 2e-16 ***
## TH 0.352388 0.057488 6.130 8.80e-10 ***
## CD 0.253930 0.084345 3.011 0.00261 **
## BA 0.286579 0.115433 2.483 0.01304 *
## CC 0.042077 0.007171 5.868 4.42e-09 ***
## TH:CD -0.026586 0.006452 -4.121 3.78e-05 ***
## BA:CC -0.006998 0.002146 -3.262 0.00111 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1163.20 on 912 degrees of freedom
## Residual deviance: 782.96 on 906 degrees of freedom
## (2 observations deleted due to missingness)
## AIC: 796.96
##
## Number of Fisher Scoring iterations: 5🌳 Some interaction terms were significant, suggesting that a combination of taller trees with deeper canopy might increase nesting likelihood.
Species-Specific Analysis
I ran stratified models to explore whether nesting predictors vary by tree species (TS):
levels(data$TS)## NULLby_species <- split(data, data$TS)
models_by_species <- lapply(by_species, function(df) {
glm(`P/A` ~ TH + CD + CC + BA + DS, data = df, family = binomial)
})## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred## Warning: glm.fit: algorithm did not converge## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred## Warning: glm.fit: algorithm did not converge## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurredlapply(models_by_species, summary)## $Acacia
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.341709 1.636531 -0.820 0.412
## TH -0.069889 0.391637 -0.178 0.858
## CD -0.051594 0.413620 -0.125 0.901
## CC 0.053601 0.040191 1.334 0.182
## BA 0.639031 0.872354 0.733 0.464
## DS -0.006525 0.022984 -0.284 0.776
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 30.316 on 21 degrees of freedom
## Residual deviance: 23.194 on 16 degrees of freedom
## AIC: 35.194
##
## Number of Fisher Scoring iterations: 4
##
##
## $Combretum
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -3.478e+01 2.560e+05 0 1
## TH 4.509e+00 1.370e+05 0 1
## CD -5.106e+00 9.998e+04 0 1
## CC 2.497e-01 1.988e+04 0 1
## BA 1.077e+00 2.563e+05 0 1
## DS 5.009e-02 3.648e+03 0 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 7.3479e+00 on 14 degrees of freedom
## Residual deviance: 2.8238e-10 on 9 degrees of freedom
## AIC: 12
##
## Number of Fisher Scoring iterations: 24
##
##
## $Comiphora
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients: (5 not defined because of singularities)
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -22.57 48196.14 0 1
## TH NA NA NA NA
## CD NA NA NA NA
## CC NA NA NA NA
## BA NA NA NA NA
## DS NA NA NA NA
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 0 degrees of freedom
## Residual deviance: 3.1675e-10 on 0 degrees of freedom
## AIC: 2
##
## Number of Fisher Scoring iterations: 21
##
##
## $`Crocodile bark`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -98.836 325354.505 0.000 1.000
## TH 30.738 153280.074 0.000 1.000
## CD -12.008 156636.425 0.000 1.000
## CC 1.965 6435.641 0.000 1.000
## BA -144.175 215198.814 -0.001 0.999
## DS -1.778 5339.491 0.000 1.000
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 2.0862e+01 on 21 degrees of freedom
## Residual deviance: 2.3929e-09 on 16 degrees of freedom
## AIC: 12
##
## Number of Fisher Scoring iterations: 25
##
##
## $`Crocodile Bark`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.457e+01 1.145e+06 0 1
## TH -1.020e-13 1.077e+06 0 1
## CD 9.925e-14 9.871e+05 0 1
## CC -2.251e-16 2.018e+04 0 1
## BA 4.714e-14 7.290e+05 0 1
## DS -4.803e-16 4.578e+03 0 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 6 degrees of freedom
## Residual deviance: 3.0007e-10 on 1 degrees of freedom
## AIC: 12
##
## Number of Fisher Scoring iterations: 23
##
##
## $`Croton Megalabotis`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients: (3 not defined because of singularities)
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.357e+01 7.648e+05 0 1
## TH 1.153e-13 3.512e+05 0 1
## CD -9.391e-14 2.924e+05 0 1
## CC NA NA NA NA
## BA NA NA NA NA
## DS NA NA NA NA
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 2 degrees of freedom
## Residual deviance: 3.4957e-10 on 0 degrees of freedom
## AIC: 6
##
## Number of Fisher Scoring iterations: 22
##
##
## $`D. mespiliformus`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients: (5 not defined because of singularities)
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -22.57 48196.14 0 1
## TH NA NA NA NA
## CD NA NA NA NA
## CC NA NA NA NA
## BA NA NA NA NA
## DS NA NA NA NA
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 0 degrees of freedom
## Residual deviance: 3.1675e-10 on 0 degrees of freedom
## AIC: 2
##
## Number of Fisher Scoring iterations: 21
##
##
## $`D. quiloensis`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 1.106e+02 2.792e+05 0 1
## TH -1.628e+01 1.009e+05 0 1
## CD 5.875e+00 1.129e+05 0 1
## CC -5.322e+00 7.849e+04 0 1
## BA -2.717e+02 2.413e+06 0 1
## DS -6.999e-01 7.469e+03 0 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1.3003e+01 on 19 degrees of freedom
## Residual deviance: 7.0792e-10 on 14 degrees of freedom
## AIC: 12
##
## Number of Fisher Scoring iterations: 25
##
##
## $`E. zambesiacum`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients: (4 not defined because of singularities)
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.357e+01 3.105e+05 0 1
## TH 1.149e-14 1.060e+05 0 1
## CD NA NA NA NA
## CC NA NA NA NA
## BA NA NA NA NA
## DS NA NA NA NA
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 1 degrees of freedom
## Residual deviance: 2.3305e-10 on 0 degrees of freedom
## AIC: 4
##
## Number of Fisher Scoring iterations: 22
##
##
## $`Monkey Orange`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients: (5 not defined because of singularities)
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -22.57 48196.14 0 1
## TH NA NA NA NA
## CD NA NA NA NA
## CC NA NA NA NA
## BA NA NA NA NA
## DS NA NA NA NA
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 0 degrees of freedom
## Residual deviance: 3.1675e-10 on 0 degrees of freedom
## AIC: 2
##
## Number of Fisher Scoring iterations: 21
##
##
## $Mopane
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.550079 0.407576 -3.803 0.000143 ***
## TH 0.146310 0.050073 2.922 0.003478 **
## CD -0.032853 0.046821 -0.702 0.482885
## CC 0.021426 0.005827 3.677 0.000236 ***
## BA 0.100625 0.086038 1.170 0.242185
## DS -0.025620 0.004876 -5.254 1.49e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 999.26 on 768 degrees of freedom
## Residual deviance: 670.93 on 763 degrees of freedom
## (2 observations deleted due to missingness)
## AIC: 682.93
##
## Number of Fisher Scoring iterations: 5
##
##
## $`P. violacea`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients: (3 not defined because of singularities)
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.357e+01 4.008e+05 0 1
## TH -1.994e-15 2.170e+04 0 1
## CD 1.331e-15 2.687e+04 0 1
## CC NA NA NA NA
## BA NA NA NA NA
## DS NA NA NA NA
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 2 degrees of freedom
## Residual deviance: 3.4957e-10 on 0 degrees of freedom
## AIC: 6
##
## Number of Fisher Scoring iterations: 22
##
##
## $Strychnos
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.457e+01 9.933e+05 0 1
## TH -8.149e-15 2.102e+05 0 1
## CD 1.471e-14 1.716e+05 0 1
## CC -3.402e-15 3.632e+04 0 1
## BA 1.280e-13 9.784e+05 0 1
## DS 7.923e-16 9.036e+03 0 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 6 degrees of freedom
## Residual deviance: 3.0007e-10 on 1 degrees of freedom
## AIC: 12
##
## Number of Fisher Scoring iterations: 23
##
##
## $Terminalia
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients: (5 not defined because of singularities)
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -22.57 48196.14 0 1
## TH NA NA NA NA
## CD NA NA NA NA
## CC NA NA NA NA
## BA NA NA NA NA
## DS NA NA NA NA
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 0 degrees of freedom
## Residual deviance: 3.1675e-10 on 0 degrees of freedom
## AIC: 2
##
## Number of Fisher Scoring iterations: 21
##
##
## $`Terminalia Maprunoid`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.457e+01 2.973e+05 0 1
## TH -1.803e-15 4.941e+04 0 1
## CD -2.230e-16 7.908e+04 0 1
## CC 2.017e-16 3.835e+04 0 1
## BA 2.367e-14 3.949e+05 0 1
## DS -6.705e-17 3.482e+03 0 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 7 degrees of freedom
## Residual deviance: 3.4294e-10 on 2 degrees of freedom
## AIC: 12
##
## Number of Fisher Scoring iterations: 23
##
##
## $`V. lanciflora`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients: (5 not defined because of singularities)
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -22.57 48196.14 0 1
## TH NA NA NA NA
## CD NA NA NA NA
## CC NA NA NA NA
## BA NA NA NA NA
## DS NA NA NA NA
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 0 degrees of freedom
## Residual deviance: 3.1675e-10 on 0 degrees of freedom
## AIC: 2
##
## Number of Fisher Scoring iterations: 21
##
##
## $`Vitex payos`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -212.724 793958.349 0 1
## TH 113.077 365934.280 0 1
## CD -128.447 450042.819 0 1
## CC 10.224 41549.781 0 1
## BA 31.464 277579.859 0 1
## DS 1.297 8597.725 0 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 7.6382e+00 on 5 degrees of freedom
## Residual deviance: 2.5720e-10 on 0 degrees of freedom
## AIC: 12
##
## Number of Fisher Scoring iterations: 23
##
##
## $`Wine Cup Tree`
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients: (2 not defined because of singularities)
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.457e+01 3.741e+05 0 1
## TH -7.635e-15 7.990e+05 0 1
## CD 7.635e-15 8.495e+05 0 1
## CC NA NA NA NA
## BA -1.136e-30 7.219e+05 0 1
## DS NA NA NA NA
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 0.0000e+00 on 4 degrees of freedom
## Residual deviance: 2.1434e-10 on 1 degrees of freedom
## AIC: 8
##
## Number of Fisher Scoring iterations: 23
##
##
## $Z.Coca
##
## Call:
## glm(formula = `P/A` ~ TH + CD + CC + BA + DS, family = binomial,
## data = df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.78286 2.67927 -0.665 0.5058
## TH 5.53253 3.23191 1.712 0.0869 .
## CD -5.46988 3.24594 -1.685 0.0920 .
## CC -0.26523 0.17907 -1.481 0.1386
## BA 3.33066 2.67379 1.246 0.2129
## DS -0.07489 0.05154 -1.453 0.1462
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 26.287 on 18 degrees of freedom
## Residual deviance: 13.559 on 13 degrees of freedom
## AIC: 25.559
##
## Number of Fisher Scoring iterations: 7Overal Model (All Specifies combined)
Key Findings:
TH (Tree Height): Positive and significant (OR = 1.16, p = 0.001). Taller trees are more likely to be present.
CC (Canopy Cover): Positive and highly significant (OR = 1.02, p < 0.001). Higher canopy cover increases presence probability.
DS (Distance to Settlement): Negative and highly significant (OR = 0.97, p < 0.001). Greater distances reduce tree presence likelihood.
Interpretation: This suggests that certain trees favor closer proximity to settlements (possibly due to less competition or anthropogenic influence), thrive in taller form, and prefer denser canopy conditions. This could point to selective growth or management in semi-natural landscapes.
2. Interaction Model: THCD and BACC
Key Findings: TH and CD: Individually positive; their interaction (TH:CD) is negative and significant (p < 0.001).
BA and CC: Individually positive; their interaction (BA:CC) is significantly negative.
Interpretation: The TH:CD interaction indicates diminishing returns: taller trees with wider canopies may not always be more likely to occur — perhaps due to crowding, wind exposure, or ecological trade-offs.
The BA:CC interaction implies that trees with both high basal area and high canopy cover may reduce each other’s individual positive effects. Possibly, only certain species can support both without resource strain.
🌳 Species-Specific Models
TH: Positive and significant (p = 0.001)
CC: Positive and significant (p < 0.001)
DS: Negative and highly significant (p < 0.001)
CD and BA: Not significant
Interpretation: Mopane presence is associated with taller trees, denser canopy, and closer proximity to settlements. It is possibly a dominant species benefiting from human land-use or less susceptible to edge effects.
🌳 Acacia None of the predictors were statistically significant.
AIC and deviance suggest acceptable fit but low explanatory power.
Interpretation: The model for Acacia shows weak relationships with the environmental variables. This might suggest:
Acacia distribution is driven by other unmeasured factors (e.g., soil chemistry, grazing).
Or the data sample size for Acacia is too small (only 22 observations) to detect significance.
🌳 Combretum, Comiphora, Crocodile Bark Common Observations: Models failed to converge or had coefficients with massive standard errors.
Warnings: “fitted probabilities numerically 0 or 1 occurred”, “did not converge”.
Interpretation: Extreme separation or sparse data: Possibly too few presences or absences for reliable modeling.
Multicollinearity or perfect prediction: Some variables might perfectly predict the outcome, leading to infinite coefficients.
Recommendation: Consider reducing model complexity (e.g., remove interactions or collinear variables).
Possibly use Firth’s penalized logistic regression for rare events.
Consider combining rare species or conducting a Poisson/Zero-Inflated model if dealing with count-based or highly zero-inflated data.
predicted probabilities
predicted <- predict(model, type = "response")
plot(data$TH, main = "Probability vs Tree Height", xlab = "Tree Height", ylab = "Predicted Probability")
📊 Model Diagnostics & Suggestions 1. Model Fit Overall model AIC = 785.8 vs Interaction Model AIC = 796.96 → Suggests simpler model may be better.
Mopane model had the best predictive power due to significant variables and large sample size.
- Model Issues Many warnings in species-specific models point to the need for:
Better sampling across species
Variable selection or transformation
Potential hierarchical models (e.g., GLMM with species as a random effect)
📌 Final Thoughts and Recommendations
| Model | Key Drivers | Notes |
|---|---|---|
| Overall | TH↑, CC↑, DS↓ | Strong predictors, interpretable |
| Mopane | Same as overall | Dominant species, reliable fit |
| Acacia | None | Weak signal, needs more data |
| Others | Not interpretable | Sparse data, consider alternative models |