I have a question about boostrapping confidence intervals for the random effects (BLUPs) of a multilevel model.
I'm currently using bootMer and there is an argument use.u=TRUE that allows one to treat the BLUPs as fixed instead of re-estimating them. Since the BLUPs are random variables it would seem appropriate to re-estimate them at each bootstrap, and indeed the default option is use.u=FALSE.
However the underlying assumption is that my clusters are a random sample of clusters drawn from a population of clusters. In my case I am running a survey experiment in 26 countries (this is the cluster of interest) which in reality were not randomly drawn. And while I am interested in drawing inferences about the larger population of countries from which my sample is drawn, I am also interested in the cluster specific effects, AKA the BLUPs, for each one of these clusters. Because of this I'm resorting to performing bootstrap to get valid confidence intervals for these "estimates".
In this case would it be OK to set use.u=TRUE?
A related question was asked here: https://stats.stackexchange.com/questions/417518/how-to-get-confidence-intervals-for-modeled-data-of-lmer-model-in-r-with-bootmer
however I'm not sure if the answer travelled to my case. Anyone have ideas?
Related
Working with a dataset of ~200 observations and a number of variables. Unfortunately, none of the variables are distributed normally. If it possible to extract a data subset where at least one desired variable will be distributed normally? Want to do some statistics after (at least logistic regression).
Any help will be much appreciated,
Phil
If there are just a few observations that skew the distribution of individual variables, and no other reasons speaking against using a particular method (such as logistic regression) on your data, you might want to study the nature of "weird" observations before deciding on which analysis method to use eventually.
I would:
carry out the desired regression analysis (e.g. logistic regression), and as it's always required, carry out residual analysis (Q-Q Normal plot, Tukey-Anscombe plot, Leverage plot, also see here) to check the model assumptions. See whether the residuals are normally distributed (the normal distribution of model residuals is the actual assumption in linear regression, not that each variable is normally distributed, of course you might have e.g. bimodally distributed data if there are differences between groups), see if there are observations which could be regarded as outliers, study them (see e.g. here), and if possible remove them from the final dataset before re-fitting the linear model without outliers.
However, you always have to state which observations were removed, and on what grounds. Maybe the outliers can be explained as errors in data collection?
The issue of whether it's a good idea to remove outliers, or a better idea to use robust methods was discussed here.
as suggested by GuedesBF, you may want to find a test or model method which has no assumption of normality.
Before modelling anything or removing any data, I would always plot the data by treatment / outcome groups, and inspect the presence of missing values. After quickly looking at your dataset, it seems that quite some variables have high levels of missingness, and your variable 15 has a lot of zeros. This can be quite problematic for e.g. linear regression.
Understanding and describing your data in a model-free way (with clever plots, e.g. using ggplot2 and multiple aesthetics) is much better than fitting a model and interpreting p-values when violating model assumptions.
A good start to get an overview of all data, their distribution and pairwise correlation (and if you don't have more than around 20 variables) is to use the psych library and pairs.panels.
dat <- read.delim("~/Downloads/dput.txt", header = F)
library(psych)
psych::pairs.panels(dat[,1:12])
psych::pairs.panels(dat[,13:23])
You can then quickly see the distribution of each variable, and the presence of correlations among each pair of variables. You can tune arguments of that function to use different correlation methods, and different displays. Happy exploratory data analysis :)
I'm trying to figure out if I'm writing my linear mixed effect model in the correct manner to compare the models.
The experimental set up:
We are looking at the length~weight relationship of a species of snails. We chose 3 sites (rivers) for these experiments that we consider a fixed effect- we expect there to be a difference, we just want to see the magnitude of that difference between rivers. We also have the snails in different individual groups (6 groups/river) so that we could keep them in a constant area to be measured. We don't expect/aren't interested in a group effect, but want to make sure it isn't causing any issues. So here we want to see the effect of the river on the length~weight relationship with cage as a source of random effects.
I have written the models as follows to compare with AIC:
No group effect:
Model1<-aov(Weight~Length*River,data=SnailData)
Group effect:
Model2<-lmer(Weight~Length*River +(1|group),data=SnailData)
Would this be correct, or is there a different way I should be looking at the group random effect?
I have a data set called Data, with 30 scaled and centered features and 1 outcome with column name OUTCOME, referred to 700k records, stored in data.table format. I computed its PCA, and observed that its first 8 components account for the 95% of the variance. I want to train a random forest in h2o, so this is what I do:
Data.pca=prcomp(Data,retx=TRUE) # compute the PCA of Data
Data.rotated=as.data.table(Data.pca$x)[,c(1:8)] # keep only first 8 components
Data.dump=cbind(Data.rotated,subset(Data,select=c(OUTCOME))) # PCA dataset plus outcomes for training
This way I have a dataset Data.dump where I have 8 features that are rotated on the PCA components, and at each record I associated its outcome.
First question: is this rational? or do I have to permute somehow the outcomes vector? or the two things are unrelated?
Then I split Data.dump in two sets, Data.train for training and Data.test for testing, all as.h2o. The I feed them to a random forest:
rf=h2o.randomForest(training_frame=Data.train,x=1:8,y=9,stopping_rounds=2,
ntrees=200,score_each_iteration=T,seed=1000000)
rf.pred=as.data.table(h2o.predict(rf,Data.test))
What happens is that rf.pred seems not so similar to the original outcomes Data.test$OUTCOME. I tried to train a neural network as well, and did not even converge, crashing R.
Second question: is it because I am carrying on some mistake from the PCA treatment? or because I badly set up the random forest? Or I am just dealing with annoying data?
I do not know where to start, as I am new to data science, but the workflow seems correct to me.
Thanks a lot in advance.
The answer to your second question (i.e. "is it the data, or did I do something wrong") is hard to know. This is why you should always try to make a baseline model first, so you have an idea of how learnable the data is.
The baseline could be h2o.glm(), and/or it could be h2o.randomForest(), but either way without the PCA step. (You didn't say if you are doing a regression or a classification, i.e. if OUTCOME is a number or a factor, but both glm and random forest will work either way.)
Going to your first question: yes, it is a reasonable thing to do, and no you don't have to (in fact, should not) involve the outcomes vector.
Another way to answer your first question is: no, it unreasonable. It may be that a random forest can see all the relations itself without needing you to use a PCA. Remember when you use a PCA to reduce the number of input dimensions you are also throwing away a bit of signal, too. You said that the 8 components only capture 95% of the variance. So you are throwing away some signal in return for having fewer inputs, which means you are optimizing for complexity at the expense of prediction quality.
By the way, concatenating the original inputs and your 8 PCA components, is another approach: you might get a better model by giving it this hint about the data. (But you might not, which is why getting some baseline models first is essential, before trying these more exotic ideas.)
I’m trying to do an ANCOVA here ...
I want to analyze the effect of EROSION FORCE and ZONATION on all the species (listed with small letters) in each POOL.STEP (ranging from 1-12/1-4), while controlling for the effect of FISH.
I’m not sure if I’m doing it right. What is the command for ANCOVA?
So far I used lm(EROSIONFORCE~ZONATION+FISH,data=d), which yields:
So what I see here is that both erosion force percentage (intercept?) and sublittoral zonation are significant in some way, but I’m still not sure if I’ve done an ANCOVA correctly here or is this just an ANOVA?
In general, ANCOVA (analysis of covariance) is simply a special case of the general linear model with one categorical predictor (factor) and one continuous predictor (the "covariate"), so lm() is the right function to use.
However ... the bottom line is that you have a moderately challenging statistical problem here, and I would strongly recommend that you try to get local help (if you're working within a research group, can you consult with others in your group about appropriate methods?) I would suggest following up either on CrossValidated or r-sig-ecology#r-project.org
by putting EROSIONFORCE on the left side of the formula, you're specifying that you want to use EROSIONFORCE as a response (dependent) variable, i.e. your model is estimating how erosion force varies across zones and for different fish numbers - nothing about species response
if you want to analyze the response of a single species to erosion and zone, controlling for fish numbers, you need something like
lm(`Acmaeidae s...` ~ EROSIONFORCE+ZONATION+FISH, data=your_data)
the lm() suggestion above would do each species independently, i.e. you'd have to do a separate analysis for each species. If you also want to do it separately for each POOL.STEP you're going to have to do a lot of separate analyses. There are various ways of automating this in R, the most idiomatic is probably to melt your data (see reshape2::melt or tidy::gather) into long format and then use lmList from lme4.
since you have count data with low means, i.e. lots of zeros (and a few big values), you should probably consider a Poisson or negative binomial model, and possibly even a zero-inflated/hurdle model (i.e. analyze presence-absence and size of positive responses separately)
if you really want to analyze the joint distribution of all species (i.e., a response of a multivariate analysis, which is the M in MANOVA), you're going to have to work quite a bit harder ... there are a variety of joint species distribution models by people like Pierre Legendre, David Warton and others ... I'd suggest you try starting with the mvabund package, but you might need to do some reading first
I am comparing various predictive models on a binary classification task using the caret R package with respect to their predictive performance (liftChart) and prediction accuracy (calibration plot). I found the following issues:
1. Sometimes the lift function is very very slow when the number of observation is quite big or there are various competing classifiers. In addition I wonder whether it is possible to manually define the cuts of the calibration plot. I have a severe imbalanced model (average probability is 5%) and the calibration plot function assumes evenly spaced cuts.
The lift plot does the calculation for every unique probability value (much like an ROC curve), which is why it is slow.
Neither of those options are available right now. You can add two issues to the github page. I'm fairly swamped right now but those shouldn't be a big deal to change (you could always contribute solutions too).
Max