Related
I am trying to evaluate themodel fit of several regressions in R, and I have run into a problem I have had multiple times now: the log-likelihood of my Poisson regression is infinite.
I'm using a non-integer dependent variable (Note: I know what I'm doing in this regard), and I'm wondering if maybe that's the problem. However, I don't get an infinite log-likelihood when running the regression with glm.nb.
Code to reproduce the issue is below.
Edit: the problem appears to go away when I coerce the DV to integer. Any idea how to get log likelihood from Poissons with non-integer DVs?
# Input Data
so_data <- data.frame(dv = c(21.0552722691125, 24.3061351414885, 7.84658638053276,
25.0294679770848, 15.8064731063311, 10.8171744654056, 31.3008088413026,
2.26643928259238, 18.4261153345417, 5.62915828161753, 17.0691184593063,
1.11959635820499, 30.0154935602592, 23.0000809735738, 28.4389825676123,
27.7678405415711, 23.7108405071757, 23.5070651053276, 14.2534787168392,
15.2058525068363, 19.7449094187771, 2.52384709295823, 29.7081691356397,
32.4723790240354, 19.2147002673637, 61.7911384519901, 10.5687170234821,
23.9047421013736, 18.4889651451222, 13.0360878554798, 15.1752866581849,
11.5205948111817, 31.3539840929108, 31.7255952728076, 25.3034625215724,
5.00013988265465, 30.2037887018226, 1.86123112349445, 3.06932041603219,
22.6739418581257, 6.33738321053804, 24.2933951601142, 14.8634827414491,
31.8302947881089, 34.8361908525564, 1.29606416941288, 13.206844629927,
28.843579313401, 25.8024295609021, 14.4414831628722, 18.2109680632694,
14.7092063453463, 10.0738043919183, 28.4124482962025, 27.1004208775326,
1.31350378236957, 14.3009307888745, 1.32555197766214, 2.70896028922312,
3.88043749517381, 3.79492216916016, 19.4507965653633, 32.1689088941444,
2.61278585713499, 41.6955885902228, 2.13466761675063, 30.4207256294235,
24.8231524369244, 20.7605955978196, 17.2182798298094, 2.11563574288652,
12.290778250655, 0.957467139696772, 16.1775287334746))
# Run Model
p_mod <- glm(dv ~ 1, data = so_data, family = poisson(link = 'log'))
# Be Confused
logLik(p_mod)
Elaborating on #ekstroem's comment: the Poisson distribution is only supported over the non-negative integers (0, 1, ...). So, technically speaking, the probability of any non-integer value is zero -- although R does allow for a little bit of fuzz, to allow for round-off/floating-point representation issues:
> dpois(1,lambda=1)
[1] 0.3678794
> dpois(1.1,lambda=1)
[1] 0
Warning message:
In dpois(1.1, lambda = 1) : non-integer x = 1.100000
> dpois(1+1e-7,lambda=1) ## fuzz
[1] 0.3678794
It is theoretically possible to compute something like a Poisson log-likelihood for non-integer values:
my_dpois <- function(x,lambda,log=FALSE) {
LL <- -lambda+x*log(lambda)-lfactorial(x)
if (log) LL else exp(LL)
}
but I would be very careful - some quick tests with integrate suggest it integrates to 1 (after I fixed the bug in it), but I haven't checked more carefully that this is really a well-posed probability distribution. (On the other hand, some reasonable-seeming posts on CrossValidated suggest that it's not insane ...)
You say "I know what I'm doing in this regard"; can you give some more of the context? Some alternative possibilities (although this is steering into CrossValidated territory) -- the best answer depends on where your data really come from (i.e., why you have "count-like" data that are non-integer but you think should be treated as Poisson).
a quasi-Poisson model (family=quasipoisson). (R will still not give you log-likelihood or AIC values in this case, because technically they don't exist -- you're supposed to do inference on the basis of the Wald statistics of the parameters; see e.g. here for more info.)
a Gamma model (probably with a log link)
if the data started out as count data that you've scaled by some measure of effort or exposure), use an appropriate offset model ...
a generalized least-squares model (nlme::gls) with an appropriate heteroscedasticity specification
Poisson log-likelihood involves calculating log(factorial(x)) (https://www.statlect.com/fundamentals-of-statistics/Poisson-distribution-maximum-likelihood). For values larger than 30 it has to be done using Stirling's approximation formula in order to avoid exceeding the limit of computer arithmetic. Sample code in Python:
# define a likelihood function. https://www.statlect.com/fundamentals-of- statistics/Poisson-distribution-maximum-likelihood
def loglikelihood_f(lmba, x):
#Using Stirling formula to avoid calculation of factorial.
#logfactorial(n) = n*ln(n) - n
n = x.size
logfactorial = x*np.log(x+0.001) - x #np.log(factorial(x))
logfactorial[logfactorial == -inf] = 0
result =\
- np.sum(logfactorial) \
- n * lmba \
+ np.log(lmba) * np.sum(x)
return result
As an assignment I had to develop and algorithm and generate a samples for a given geometric distribution with PMF
Using the inverse transform method, I came up with the following expression for generating the values:
Where U represents a value, or n values depending on the size of the sample, drawn from a Unif(0,1) distribution and p is 0.3 as stated in the PMF above.
I have the algorithm, the implementation in R and I already generated QQ Plots to visually assess the adjustment of the empirical values to the theoretical ones (generated with R), i.e., if the generated sample follows indeed the geometric distribution.
Now I wanted to submit the generated sample to a goodness of fit test, namely the Chi-square, yet I'm having trouble doing this in R.
[I think this was moved a little hastily, in spite of your response to whuber's question, since I think before solving the 'how do I write this algorithm in R' problem, it's probably more important to deal with the 'what you're doing is not the best approach to your problem' issue (which certainly belongs where you posted it). Since it's here, I will deal with the 'doing it in R' aspect, but I would urge to you go back an ask about the second question (as a new post).]
Firstly the chi-square test is a little different depending on whether you test
H0: the data come from a geometric distribution with parameter p
or
H0: the data come from a geometric distribution with parameter 0.3
If you want the second, it's quite straightforward. First, with the geometric, if you want to use the chi-square approximation to the distribution of the test statistic, you will need to group adjacent cells in the tail. The 'usual' rule - much too conservative - suggests that you need an expected count in every bin of at least 5.
I'll assume you have a nice large sample size. In that case, you'll have many bins with substantial expected counts and you don't need to worry so much about keeping it so high, but you will still need to choose how you will bin the tail (whether you just choose a single cut-off above which all values are grouped, for example).
I'll proceed as if n were say 1000 (though if you're testing your geometric random number generation, that's pretty low).
First, compute your expected counts:
dgeom(0:20,.3)*1000
[1] 300.0000000 210.0000000 147.0000000 102.9000000 72.0300000 50.4210000
[7] 35.2947000 24.7062900 17.2944030 12.1060821 8.4742575 5.9319802
[13] 4.1523862 2.9066703 2.0346692 1.4242685 0.9969879 0.6978915
[19] 0.4885241 0.3419669 0.2393768
Warning, dgeom and friends goes from x=0, not x=1; while you can shift the inputs and outputs to the R functions, it's much easier if you subtract 1 from all your geometric values and test that. I will proceed as if your sample has had 1 subtracted so that it goes from 0.
I'll cut that off at the 15th term (x=14), and group 15+ into its own group (a single group in this case). If you wanted to follow the 'greater than five' rule of thumb, you'd cut it off after the 12th term (x=11). In some cases (such as smaller p), you might want to split the tail across several bins rather than one.
> expec <- dgeom(0:14,.3)*1000
> expec <- c(expec, 1000-sum(expec))
> expec
[1] 300.000000 210.000000 147.000000 102.900000 72.030000 50.421000
[7] 35.294700 24.706290 17.294403 12.106082 8.474257 5.931980
[13] 4.152386 2.906670 2.034669 4.747562
The last cell is the "15+" category. We also need the probabilities.
Now we don't yet have a sample; I'll just generate one:
y <- rgeom(1000,0.3)
but now we want a table of observed counts:
(x <- table(factor(y,levels=0:14),exclude=NULL))
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 <NA>
292 203 150 96 79 59 47 25 16 10 6 7 0 2 5 3
Now you could compute the chi-square directly and then calculate the p-value:
> (chisqstat <- sum((x-expec)^2/expec))
[1] 17.76835
(pval <- pchisq(chisqstat,15,lower.tail=FALSE))
[1] 0.2750401
but you can also get R to do it:
> chisq.test(x,p=expec/1000)
Chi-squared test for given probabilities
data: x
X-squared = 17.7683, df = 15, p-value = 0.275
Warning message:
In chisq.test(x, p = expec/1000) :
Chi-squared approximation may be incorrect
Now the case for unspecified p is similar, but (to my knowledge) you can no longer get chisq.test to do it directly, you have to do it the first way, but you have to estimate the parameter from the data (by maximum likelihood or minimum chi-square), and then test as above but you have one fewer degree of freedom for estimating the parameter.
See the example of doing a chi-square for a Poisson with estimated parameter here; the geometric follows the much same approach as above, with the adjustments as at the link (dealing with the unknown parameter, including the loss of 1 degree of freedom).
Let us assume you've got your randomly-generated variates in a vector x. You can do the following:
x <- rgeom(1000,0.2)
x_tbl <- table(x)
x_val <- as.numeric(names(x_tbl))
x_df <- data.frame(count=as.numeric(x_tbl), value=x_val)
# Expand to fill in "gaps" in the values caused by 0 counts
all_x_val <- data.frame(value = 0:max(x_val))
x_df <- merge(all_x_val, x_df, by="value", all.x=TRUE)
x_df$count[is.na(x_df$count)] <- 0
# Get theoretical probabilities
x_df$eprob <- dgeom(x_df$val, 0.2)
# Chi-square test: once with asymptotic dist'n,
# once with bootstrap evaluation of chi-sq test statistic
chisq.test(x=x_df$count, p=x_df$eprob, rescale.p=TRUE)
chisq.test(x=x_df$count, p=x_df$eprob, rescale.p=TRUE,
simulate.p.value=TRUE, B=10000)
There's a "goodfit" function described as "Goodness-of-fit Tests for Discrete Data" in package "vcd".
G.fit <- goodfit(x, type = "nbinomial", par = list(size = 1))
I was going to use the code you had posted in an earlier question, but it now appears that you have deleted that code. I find that offensive. Are you using this forum to gather homework answers and then defacing it to remove the evidence? (Deleted questions can still be seen by those of us with sufficient rep, and the interface prevents deletion of question with upvoted answers so you should not be able to delete this one.)
Generate a QQ Plot for testing a geometrically distributed sample
--- question---
I have a sample of n elements generated in R with
sim.geometric <- function(nvals)
{
p <- 0.3
u <- runif(nvals)
ceiling(log(u)/log(1-p))
}
for which i want to test its distribution, specifically if it indeed follows a geometric distribution. I want to generate a QQ PLot but have no idea how to.
--------reposted answer----------
A QQ-plot should be a straight line when compared to a "true" sample drawn from a geometric distribution with the same probability parameter. One gives two vectors to the functions which essentially compares their inverse ECDF's at each quantile. (Your attempt is not particularly successful:)
sim.res <- sim.geometric(100)
sim.rgeom <- rgeom(100, 0.3)
qqplot(sim.res, sim.rgeom)
Here I follow the lead of the authors of qqplot's help page (which results in flipping that upper curve around the line of identity):
png("QQ.png")
qqplot(qgeom(ppoints(100),prob=0.3), sim.res,
main = expression("Q-Q plot for" ~~ {G}[n == 100]))
dev.off()
---image not included---
You can add a "line of good fit" by plotting a line through through the 25th and 75th percentile points for each distribution. (I added a jittering feature to this to get a better idea where the "probability mass" was located:)
sim.res <- sim.geometric(500)
qqplot(jitter(qgeom(ppoints(500),prob=0.3)), jitter(sim.res),
main = expression("Q-Q plot for" ~~ {G}[n == 100]), ylim=c(0,max( qgeom(ppoints(500),prob=0.3),sim.res )),
xlim=c(0,max( qgeom(ppoints(500),prob=0.3),sim.res )))
qqline(sim.res, distribution = function(p) qgeom(p, 0.3),
prob = c(0.25, 0.75), col = "red")
I need to do some robust data-fitting operation.
I have bunch of (x,y) data, that I want to fit to a Gaussian (aka normal) function.
The point is, I want to remove the ouliers. As one can see on the sample plot below, there is another distribution of data thats pollutting my data on the right, and I don't want to take it into account to do the fitting (i.e. to find \sigma, \mu and the overall scale parameter).
R seems to be the right tool for the job, I found some packages (robust, robustbase, MASS for example) that are related to robust fitting.
However, they assume the user already has a strong knowledge of R, which is not my case, and the documentation is only provided as a sort of reference manual, no tutorial or equivalent. My statistical background is rather low, I attempted to read reference material on fitting with R, but it didn't really help (and I'm not even sure thats the right way to go).
But I have the feeling that this is actually a quite simple operation.
I have checked this related question (and the linked ones), however they take as input a single vector of values, and I have a vector of pairs, so I don't see how to transpose.
Any help on how to do this would be appreciated.
Fitting a Gaussian curve to the data, the principle is to minimise the sum of squares difference between the fitted curve and the data, so we define f our objective function and run optim on it:
fitG =
function(x,y,mu,sig,scale){
f = function(p){
d = p[3]*dnorm(x,mean=p[1],sd=p[2])
sum((d-y)^2)
}
optim(c(mu,sig,scale),f)
}
Now, extend this to two Gaussians:
fit2G <- function(x,y,mu1,sig1,scale1,mu2,sig2,scale2,...){
f = function(p){
d = p[3]*dnorm(x,mean=p[1],sd=p[2]) + p[6]*dnorm(x,mean=p[4],sd=p[5])
sum((d-y)^2)
}
optim(c(mu1,sig1,scale1,mu2,sig2,scale2),f,...)
}
Fit with initial params from the first fit, and an eyeballed guess of the second peak. Need to increase the max iterations:
> fit2P = fit2G(data$V3,data$V6,6,.6,.02,8.3,0.10,.002,control=list(maxit=10000))
Warning messages:
1: In dnorm(x, mean = p[1], sd = p[2]) : NaNs produced
2: In dnorm(x, mean = p[4], sd = p[5]) : NaNs produced
3: In dnorm(x, mean = p[4], sd = p[5]) : NaNs produced
> fit2P
$par
[1] 6.035610393 0.653149616 0.023744876 8.317215066 0.107767881 0.002055287
What does this all look like?
> plot(data$V3,data$V6)
> p = fit2P$par
> lines(data$V3,p[3]*dnorm(data$V3,p[1],p[2]))
> lines(data$V3,p[6]*dnorm(data$V3,p[4],p[5]),col=2)
However I would be wary about statistical inference about your function parameters...
The warning messages produced are probably due to the sd parameter going negative. You can fix this and also get a quicker convergence by using L-BFGS-B and setting a lower bound:
> fit2P = fit2G(data$V3,data$V6,6,.6,.02,8.3,0.10,.002,control=list(maxit=10000),method="L-BFGS-B",lower=c(0,0,0,0,0,0))
> fit2P
$par
[1] 6.03564202 0.65302676 0.02374196 8.31424025 0.11117534 0.00208724
As pointed out, sensitivity to initial values is always a problem with curve fitting things like this.
Fitting a Gaussian:
# your data
set.seed(0)
data <- c(rnorm(100,0,1), 10, 11)
# find & remove outliers
outliers <- boxplot(data)$out
data <- setdiff(data, outliers)
# fitting a Gaussian
mu <- mean(data)
sigma <- sd(data)
# testing the fit, check the p-value
reference.data <- rnorm(length(data), mu, sigma)
ks.test(reference.data, data)
I have a series of data, these are obtained through a molecular dynamics simulation, and therefore are sequential in time and correlated to some extent. I can calculate the mean as the average of the data, I want to estimate the the error associated to mean calculated in this way.
According to this book I need to calculate the "statistical inefficiency", or roughly the correlation time for the data in the series. For this I have to divide the series in blocks of varying length and, for each block length (t_b), the variance of the block averages (v_b). Then, if the variance of the whole series is v_a (that is, v_b when t_b=1), I have to obtain the limit, as t_b tends to infinity, of (t_b*v_b/v_a), and that is the inefficiency s.
Then the error in the mean is sqrt(v_a*s/N), where N is the total number of points. So, this means that only one every s points is uncorrelated.
I assume this can be done with R, and maybe there's some package that does it already, but I'm new to R. Can anyone tell me how to do it? I have already found out how to read the data series and calculate the mean and variance.
A data sample, as requested:
# t(ps) dH/dl(kJ/mol)
0.0000 582.228
0.0100 564.735
0.0200 569.055
0.0300 549.917
0.0400 546.697
0.0500 548.909
0.0600 567.297
0.0700 638.917
0.0800 707.283
0.0900 703.356
0.1000 685.474
0.1100 678.07
0.1200 687.718
0.1300 656.729
0.1400 628.763
0.1500 660.771
0.1600 663.446
0.1700 637.967
0.1800 615.503
0.1900 605.887
0.2000 618.627
0.2100 587.309
0.2200 458.355
0.2300 459.002
0.2400 577.784
0.2500 545.657
0.2600 478.857
0.2700 533.303
0.2800 576.064
0.2900 558.402
0.3000 548.072
... and this goes on until 500 ps. Of course, the data I need to analyze is the second column.
Suppose x is holding the sequence of data (e.g., data from your second column).
v = var(x)
m = mean(x)
n = length(x)
si = c()
for (t in seq(2, 1000)) {
nblocks = floor(n/t)
xg = split(x[1:(nblocks*t)], factor(rep(1:nblocks, rep(t, nblocks))))
v2 = sum((sapply(xg, mean) - m)**2)/nblocks
#v rather than v1
si = c(si, t*v2/v)
}
plot(si)
Below image is what I got from some of my time series data. You have your lower limit of t_b when the curve of si becomes approximately flat (slope = 0). See http://dx.doi.org/10.1063/1.1638996 as well.
There are a couple different ways to calculate the statistical inefficiency, or integrated autocorrelation time. The easiest, in R, is with the CODA package. They have a function, effectiveSize, which gives you the effective sample size, which is the total number of samples divided by the statistical inefficiency. The asymptotic estimator for the standard deviation in the mean is sd(x)/sqrt(effectiveSize(x)).
require('coda')
n_eff = effectiveSize(x)
Well it's never too late to contribute to a question, isn't it?
As I'm doing some molecular simulation myself, I did step uppon this problem but did not see this thread already. I found out that the method actually proposed by Allen & Tildesley seems a bit out dated compared to modern error analysis methods. The rest of the book is good enought to worth the look though.
While Sunhwan Jo's answer is correct concerning block averages method,concerning error analysis you can find other methods like the jacknife and bootstrap methods (closely related to one another) here: http://www.helsinki.fi/~rummukai/lectures/montecarlo_oulu/lectures/mc_notes5.pdf
In short, with the bootstrap method, you can make series of random artificial samples from your data and calculate the value you want on your new sample. I wrote a short piece of Python code to work some data out (don't forget to import numpy or the functions I used):
def Bootstrap(data):
B = 100 # arbitraty number of artificial samplings
es = 0.
means = numpy.zeros(B)
sizeB = data.shape[0]/4 # (assuming you pass a numpy array)
# arbitrary bin-size proportional to the one of your
# sampling.
for n in range(B):
for i in range(sizeB):
# if data is multi-column array you may have to add the one you use
# specifically in randint, else it will give you a one dimension array.
# Check the doc.
means[n] = means[n] + data[numpy.random.randint(0,high=data.shape[0])] # Assuming your desired value is the mean of the values
# Any calculation is ok.
means[n] = means[n]/sizeB
es = numpy.std(means,ddof = 1)
return es
I know it can be upgraded but it's a first shot. With your data, I get the following:
Mean = 594.84368
Std = 66.48475
Statistical error = 9.99105
I hope this helps anyone stumbling across this problem in statistical analysis of data. If I'm wrong or anything else (first post and I'm no mathematician), any correction is welcomed.
I am applying guantile regression for my data-set (using R). It is easy to produce the nice scatterplot-image with different quantile regression lines
(taus <- c(0.05,0.25,0.75,0.95)).
Problem occurs when I want to produce p-values (in order to see statistical significance of each regression line) for each one of these quantiles. For median quantile (tau=0.5) this is not problematic, but when it comes to for example tau=0.25, I get following error message:
>QRmodel<-rq(y~x,tau=0.25,model=T)
>summary(QRmodel,se="nid")
Error in summary.rq(QRmodel, se = "nid") : tau - h < 0: error in summary.rq
What could be the reason for this?
Also: Is it recommendable to mention p-values and coefficients regarding the results of quantile regression model or could it be enough to show just the plot-picture and discuss the results based on that picture?
Best regards, frustrated person
A good way to learn what's going on in these sorts of debugging situations is to find the relevant portion of code that is throwing the error. If you type 'summary.rq' at the console, you'll see the code for the function summary.rq. Scanning through it you'll find the section where it calculates se's using the "nid" method, starting with this code:
else if (se == "nid") {
h <- bandwidth.rq(tau, n, hs = hs)
if (tau + h > 1)
stop("tau + h > 1: error in summary.rq")
if (tau - h < 0)
stop("tau - h < 0: error in summary.rq")
bhi <- rq.fit.fnb(x, y, tau = tau + h)$coef
blo <- rq.fit.fnb(x, y, tau = tau - h)$coef
So what's happening here is that in order to calculate the se's, the function first need to calculate a bandwidth, h, and the quantreg model is refit for tau +/- h. For tau's near 0 or 1, there's a possibility that adding or subtracting the bandwidth 'h' will lead to a tau below 0 or greater than 1, which isn't good, so the function stops.
You have a couple of options:
1.) Try a different se method (bootstrapping?)
2.) Modify the summary.rq code yourself to force it to use either max(tau,0) or min(tau,1) in the instances where the bandwidth pushes tau out of bounds. (There could be serious theoretical reasons why this is a bad idea; not advised unless you know what you're doing.)
3.) You could try to read up on the theory behind the calculation of these se's so you'd have a better idea of when they might work well or not. This might shed some light on why you're running into errors with values of tau near 0 or 1.
Try summary(QRmodel,se="boot")
Have a look at the help for summary.rq as well!