I'm trying to demonstrate that there is an important difference between two ways of making linear model predictions. The first way, which my heart tells me is more correct, uses predict.lm which as I understand preserves the correlations between coefficients. The second approach tries to use the parameters independently.
Is this the correct way to show the difference? The two approaches seem somewhat close.
Also, is the StdErr of the coefficients the same as the standard deviation of their distributions? Or have I misunderstood what the model table is saying.
Below is a quick reprex to show what I mean:
# fake dataset
xs <- runif(200, min = -1, max = 1)
true_inter <- -1.3
true_slope <- 3.1
ybar <- true_inter + true_slope*xs
ys <- rnorm(200, ybar, sd = 1)
model <- lm(ys~xs)
# predictions
coef_sterr <- summary(model)$coefficients
inters <- rnorm(500, mean = coef_sterr[1,1], sd = coef_sterr[1,2])
slopes <- rnorm(500, mean = coef_sterr[2,1], sd = coef_sterr[2,2])
newx <- seq(from = -1, to= 1, length.out = 20)
avg_predictions <- cbind(1, newx) %*% rbind(inters, slopes)
conf_predictions <- apply(avg_predictions, 1, quantile, probs = c(.25, .975), simplify = TRUE)
# from confint
conf_interval <- predict(model, newdata=data.frame(xs = newx),
interval="confidence",
level = 0.95)
# plot to visualize
plot(ys~xs)
# averages are exactly the same
abline(model)
abline(a = coef(model)[1], b = coef(model)[2], col = "red")
# from predict, using parameter covariance
matlines(newx, conf_interval[,2:3], col = "blue", lty=1, lwd = 3)
# from simulated lines, ignoring parameter covariance
matlines(newx, t(conf_predictions), col = "orange", lty = 1, lwd = 2)
Created on 2022-04-05 by the reprex package (v2.0.1)
In this case, they would be close because there is very little correlation between the model parameters, so drawing them from two independent normals versus a multivariate normal is not that different:
set.seed(519)
xs <- runif(200, min = -1, max = 1)
true_inter <- -1.3
true_slope <- 3.1
ybar <- true_inter + true_slope*xs
ys <- rnorm(200, ybar, sd = 1)
model <- lm(ys~xs)
cov2cor(vcov(model))
# (Intercept) xs
# (Intercept) 1.00000000 -0.08054106
# xs -0.08054106 1.00000000
Also, it is probably worth calculating both of the intervals the same way, though it shouldn't make that much difference. That said, 500 observations may not be enough to get reliable estimates of the 2.5th and 97.5th percentiles of the distribution. Let's consider a slightly more complex example. Here, the two X variables are correlated - the correlation of the parameters derives in part from the correlation of the columns of the design matrix, X.
set.seed(519)
X <- MASS::mvrnorm(200, c(0,0), matrix(c(1,.65,.65,1), ncol=2))
b <- c(-1.3, 3.1, 2.5)
ytrue <- cbind(1,X) %*% b
y <- ytrue + rnorm(200, 0, .5*sd(ytrue))
dat <- data.frame(y=y, x1=X[,1], x2=X[,2])
model <- lm(y ~ x1 + x2, data=dat)
cov2cor(vcov(model))
# (Intercept) x1 x2
# (Intercept) 1.00000000 0.02417386 -0.01515887
# x1 0.02417386 1.00000000 -0.73228003
# x2 -0.01515887 -0.73228003 1.00000000
In this example, the coefficients for x1 and x2 are correlated around -0.73. As you'll see, this still doesn't result in a huge difference. Let's calculate the relevant statistics.
First, we draw B1 using the multivariate method that you rightly suspect is correct. Then, we'll draw B2 from a bunch of independent normals (actually, I'm using a multivariate normal with a diagonal variance-covariance matrix, which is the same thing).
b_est <- coef(model)
v <- vcov(model)
B1 <- MASS::mvrnorm(2500, b_est, v, empirical=TRUE)
B2 <- MASS::mvrnorm(2500, b_est, diag(diag(v)), empirical = TRUE)
Now, let's make a hypothetical X matrix and generate the relevant predictions:
hypX <- data.frame(x1=seq(-3,3, length=50),
x2 = mean(dat$x2))
yhat1 <- as.matrix(cbind(1, hypX)) %*% t(B1)
yhat2 <- as.matrix(cbind(1, hypX)) %*% t(B2)
Then we can calculate confidence intervals, etc...
yh1_ci <- t(apply(yhat1, 1, function(x)unname(quantile(x, c(.025,.975)))))
yh2_ci <- t(apply(yhat2, 1, function(x)unname(quantile(x, c(.025,.975)))))
yh1_ci <- as.data.frame(yh1_ci)
yh2_ci <- as.data.frame(yh2_ci)
names(yh1_ci) <- names(yh2_ci) <- c("lwr", "upr")
yh1_ci$fit <- c(as.matrix(cbind(1, hypX)) %*% b_est)
yh2_ci$fit <- c(as.matrix(cbind(1, hypX)) %*% b_est)
yh1_ci$method <- factor(1, c(1,2), labels=c("Multivariate", "Independent"))
yh2_ci$method <- factor(2, c(1,2), labels=c("Multivariate", "Independent"))
yh1_ci$x1 <- hypX[,1]
yh2_ci$x1 <- hypX[,1]
yh <- rbind(yh1_ci, yh2_ci)
We could then plot the two confidence intervals as you did.
ggplot(yh, aes(x=x1, y=fit, ymin=lwr, ymax=upr, fill=method)) +
geom_ribbon(colour="transparent", alpha=.25) +
geom_line() +
theme_classic()
Perhaps a better visual would be to compare the widths of the intervals.
w1 <- yh1_ci$upr - yh1_ci$lwr
w2 <- yh2_ci$upr - yh2_ci$lwr
ggplot() +
geom_point(aes(x=hypX[,1], y=w2-w1)) +
theme_classic() +
labs(x="x1", y="Width (Independent) - Width (Multivariate)")
This shows that for small values of x1, the independent confidence intervals are wider than the multivariate ones. For values of x1 above 0, it's a more mixed bag.
This tells you that there is some difference, but you don't need the simulation to know which one is 'right'. That's because the prediction is a linear combination of constants and random variables.
In this case, the b terms are the random variables and the x values are the constants. We know that the variance of a linear combination can be calculated this way:
All that is to say that your intuition is correct.
Related
I made a plot for my data and am now I would like to have the difference in y for every x that was estimated by geom_smooth(). There is a similiar question which unfortunately has no answer. For example, how to get the differences for the following plot (data below):
EDIT
Two suggestions were made but I still don't know how to calculate the differences.
First suggestion was to access the data from the ggplot object. I did so with
pb <- ggplot_build(p)
pb[["data"]][[1]]
That approach kind of works, but the data doesn't use the same x values for the groups. For example, the first x value of the first group is -3.21318853, but there is no x of -3.21318853 for the second group, hence, I can not calculate the difference in y for -3.21318853 between both groups
Second suggestion was to see what formula is used in geom_smooth(). The package description says that "loess() is used for less than 1,000 observations; otherwise mgcv::gam() is used with formula = y ~ s(x, bs = "cs")". My N is more than 60,000, hence, gam is used by default. I am not familiar with gam; can anyone provide a short answer how to calculate the difference between the two lines considering the things just described?
R Code
library("ggplot2") # library ggplot
set.seed(1) # make example reproducible
n <- 5000 # set sample size
df <- data.frame(x= rnorm(n), g= factor(rep(c(0,1), n/2))) # generate data
df$y <- NA # include y in df
df$y[df$g== 0] <- df$x[df$g== 0]**2 + rnorm(sum(df$g== 0))*5 # y for group g= 0
df$y[df$g== 1] <-2 + df$x[df$g== 1]**2 + rnorm(sum(df$g== 1))*5 # y for g= 1 (with intercept 2)
ggplot(df, aes(x, y, col= g)) + geom_smooth() + geom_point(alpha= .1) # make a plot
Hi and welcome on Stack Overflow,
The first suggestion is good. To make the x-sequences match, you can interpolate the values in between using the approx function (in stats).
library("ggplot2") # library ggplot
set.seed(1) # make example reproducible
n <- 5000 # set sample size
df <- data.frame(x= rnorm(n), g= factor(rep(c(0,1), n/2))) # generate data
df$y <- NA # include y in df
df$y[df$g== 0] <- df$x[df$g== 0]**2 + rnorm(sum(df$g== 0))*5 # y for group g= 0
df$y[df$g== 1] <-2 + df$x[df$g== 1]**2 + rnorm(sum(df$g== 1))*5 # y for g= 1 (with intercept 2)
p <- ggplot(df, aes(x, y, col= g)) + geom_smooth() + geom_point(alpha= .1) # make a plot
pb <- ggplot_build(p) # Get computed data
data.of.g1 <- pb[['data']][[1]][pb[['data']][[1]]$group == 1, ] # Extract info for group 1
data.of.g2 <- pb[['data']][[1]][pb[['data']][[1]]$group == 2, ] # Extract info for group 2
xlimit.inf <- max(min(data.of.g1$x), min(data.of.g2$x)) # Get the minimum X the two smoothed data have in common
xlimit.sup <- min(max(data.of.g1$x), max(data.of.g2$x)) # Get the maximum X
xseq <- seq(xlimit.inf, xlimit.sup, 0.01) # Sequence of X value (you can use bigger/smaller step size)
# Based on data from group 1 and group 2, interpolates linearly for all the values in `xseq`
y.g1 <- approx(x = data.of.g1$x, y = data.of.g1$y, xout = xseq)
y.g2 <- approx(x = data.of.g2$x, y = data.of.g2$y, xout = xseq)
difference <- data.frame(x = xseq, dy = abs(y.g1$y - y.g2$y)) # Compute the difference
ggplot(difference, aes(x = x, y = dy)) + geom_line() # Make the plot
Output:
As I mentioned in the comments above, you really are better off doing this outside of ggplot and instead do it with a full model of the two smooths from which you can compute uncertainties on the difference, etc.
This is basically a short version of a blog post that I wrote a year or so back.
OP's exmaple data
set.seed(1) # make example reproducible
n <- 5000 # set sample size
df <- data.frame(x= rnorm(n), g= factor(rep(c(0,1), n/2))) # generate data
df$y <- NA # include y in df
df$y[df$g== 0] <- df$x[df$g== 0]**2 + rnorm(sum(df$g== 0))*5 # y for group g= 0
df$y[df$g== 1] <-2 + df$x[df$g== 1]**2 + rnorm(sum(df$g== 1))*5 # y for g= 1 (with intercept 2)
Start by fitting the model for the example data:
library("mgcv")
m <- gam(y ~ g + s(x, by = g), data = df, method = "REML")
Here I'm fitting a GAM with a factor-smooth interaction (the by bit) and for this model we need to also include g as a parametric effect as the group-specific smooths are both centred about 0 so we need to include the group means in the parametric part of the model.
Next we need a grid of data along the x variable at which we will estimate the difference between the two estimated smooths:
pdat <- with(df, expand.grid(x = seq(min(x), max(x), length = 200),
g = c(0,1)))
pdat <- transform(pdat, g = factor(g))
then we use this prediction data to generate the Xp matrix, which is a matrix that maps values of the covariates to values of the basis expansion for the smooths; we can manipulate this matrix to get the difference smooth that we want:
xp <- predict(m, newdata = pdat, type = "lpmatrix")
Next some code to identify which rows and columns in xp belong to the smooths for the respective levels of g; as there are only two levels and only a single smooth term in the model, this is entirely trivial but for more complex models this is needed and it is important to get the smooth component names right for the grep() bits to work.
## which cols of xp relate to splines of interest?
c1 <- grepl('g0', colnames(xp))
c2 <- grepl('g1', colnames(xp))
## which rows of xp relate to sites of interest?
r1 <- with(pdat, g == 0)
r2 <- with(pdat, g == 1)
Now we can difference the rows of xp for the pair of levels we are comparing
## difference rows of xp for data from comparison
X <- xp[r1, ] - xp[r2, ]
As we focus on the difference, we need to zero out all the column not associated with the selected pair of smooths, which includes any parametric terms.
## zero out cols of X related to splines for other lochs
X[, ! (c1 | c2)] <- 0
## zero out the parametric cols
X[, !grepl('^s\\(', colnames(xp))] <- 0
(In this example, these two lines do exactly the same thing, but in more complex examples both are needed.)
Now we have a matrix X which contains the difference between the two basis expansions for the pair of smooths we're interested in, but to get this in terms of fitted values of the response y we need to multiply this matrix by the vector of coefficients:
## difference between smooths
dif <- X %*% coef(m)
Now dif contains the difference between the two smooths.
We can use X again and covariance matrix of the model coefficients to compute the standard error of this difference and thence a 95% (in this case) confidence interval for the estimate difference.
## se of difference
se <- sqrt(rowSums((X %*% vcov(m)) * X))
## confidence interval on difference
crit <- qt(.975, df.residual(m))
upr <- dif + (crit * se)
lwr <- dif - (crit * se)
Note that here with the vcov() call we're using the empirical Bayesian covariance matrix but not the one corrected for having chosen the smoothness parameters. The function I show shortly allows you to account for this additional uncertainty via argument unconditional = TRUE.
Finally we gather the results and plot:
res <- data.frame(x = with(df, seq(min(x), max(x), length = 200)),
dif = dif, upr = upr, lwr = lwr)
ggplot(res, aes(x = x, y = dif)) +
geom_ribbon(aes(ymin = lwr, ymax = upr, x = x), alpha = 0.2) +
geom_line()
This produces
Which is consistent with an assessment that shows the model with the group-level smooths doesn't provide substantially better fit than a model with different group means but only single common smoother in x:
r$> m0 <- gam(y ~ g + s(x), data = df, method = "REML")
r$> AIC(m0, m)
df AIC
m0 9.68355 30277.93
m 14.70675 30285.02
r$> anova(m0, m, test = 'F')
Analysis of Deviance Table
Model 1: y ~ g + s(x)
Model 2: y ~ g + s(x, by = g)
Resid. Df Resid. Dev Df Deviance F Pr(>F)
1 4990.1 124372
2 4983.9 124298 6.1762 73.591 0.4781 0.8301
Wrapping up
The blog post I mentioned has a function which wraps the steps above into a simple function, smooth_diff():
smooth_diff <- function(model, newdata, f1, f2, var, alpha = 0.05,
unconditional = FALSE) {
xp <- predict(model, newdata = newdata, type = 'lpmatrix')
c1 <- grepl(f1, colnames(xp))
c2 <- grepl(f2, colnames(xp))
r1 <- newdata[[var]] == f1
r2 <- newdata[[var]] == f2
## difference rows of xp for data from comparison
X <- xp[r1, ] - xp[r2, ]
## zero out cols of X related to splines for other lochs
X[, ! (c1 | c2)] <- 0
## zero out the parametric cols
X[, !grepl('^s\\(', colnames(xp))] <- 0
dif <- X %*% coef(model)
se <- sqrt(rowSums((X %*% vcov(model, unconditional = unconditional)) * X))
crit <- qt(alpha/2, df.residual(model), lower.tail = FALSE)
upr <- dif + (crit * se)
lwr <- dif - (crit * se)
data.frame(pair = paste(f1, f2, sep = '-'),
diff = dif,
se = se,
upper = upr,
lower = lwr)
}
Using this function we can repeat the entire analysis and plot the difference with:
out <- smooth_diff(m, pdat, '0', '1', 'g')
out <- cbind(x = with(df, seq(min(x), max(x), length = 200)),
out)
ggplot(out, aes(x = x, y = diff)) +
geom_ribbon(aes(ymin = lower, ymax = upper, x = x), alpha = 0.2) +
geom_line()
I won't show the plot here as it is identical to that shown above except for the axis labels.
I want to test a model where the distribution of a random variable, assumed normal, is conditional on the regime of another random variable, that switches state according to a Markov chain. The first step would be:
Assuming the simple linear model:
lm(y~x, data=data)
I want to find the parameters of the distribution assuming that x switches regime.
For example:
mkt.bull <- rnorm(150, 2, 1.5)
mkt.bear <- rnorm(150, -1, 2.5)
x <- c(mkt.bear,mkt.bull)
portfolio.bull <- rnorm(150, 1.75, 1.6)
portfolio.bear <- rnorm(150, -0.5, 2.3)
y <- c(portfolio.bear,portfolio.bull)
In the example above, x can be modelled as a Markov switching model (msmFit) with two states, one bull and one bear. Instead of approaching the problem with a lm,
lm(y~x)
since the two series are clearly non-linear, I want to run a regression where the parameters are conditional on the regime. This can be done with maximum likelihood, but the first step is to define the distribution of y as:
y_i | x, S_t ~ N(alpha + beta_{i,s_t}); sigma^2)
How can I code the above formula? I guess this cannot be done using rnorm. Is there another way?
Thanks
Data
Here I prepared and visualized the data.
# Load packages
library(tidyverse)
library(rjags)
# Set seed for reproduciblility
set.seed(199)
mkt.bull <- rnorm(150, 2, 1.5)
mkt.bear <- rnorm(150, -1, 2.5)
x <- c(mkt.bear,mkt.bull)
portfolio.bull <- rnorm(150, 1.75, 1.6)
portfolio.bear <- rnorm(150, -0.5, 2.3)
y <- c(portfolio.bear,portfolio.bull)
# Create example data frame
dat <- data.frame(x = x, y = y, regime = c(rep("bear", 150), rep("bull", 150)),
stringsAsFactors = FALSE)
# Plot the sample distribution
dat$regime <- factor(dat$regime, levels = c("bear", "bull"))
# Create a plot
ggplot(dat, aes(x = y, color = regime)) +
geom_density()
There are two regimes, bear and bull. The y for these regimes are both normally distributed. It seems like the OP wants to estimate the mean and standard deviation of y conditioned on these states.
Maximum Likelihood
Here is one way to use maximum likelihood to estimate the parameters using the stats4 package.
# Load the infer package
library(stats4)
# Split the data
y_bull <- dat %>% filter(regime %in% "bull") %>% pull("y")
y_bear <- dat %>% filter(regime %in% "bear") %>% pull("y")
# Define the log-likelihood function
LogLike_bull <- function(Mean, Sigma){
R <- suppressWarnings(dnorm(y_bull, Mean, Sigma))
return(-sum(log(R)))
}
LogLike_bear <- function(Mean, Sigma){
R <- suppressWarnings(dnorm(y_bear, Mean, Sigma))
return(-sum(log(R)))
}
mle(minuslogl = LogLike_bull, start = list(Mean = 1, Sigma = 1))
# Call:
# mle(minuslogl = LogLike_bull, start = list(Mean = 1, Sigma = 1))
#
# Coefficients:
# Mean Sigma
# 1.703099 1.482619
mle(minuslogl = LogLike_bear, start = list(Mean = 1, Sigma = 1))
# Call:
# mle(minuslogl = LogLike_bear, start = list(Mean = 1, Sigma = 1))
#
# Coefficients:
# Mean Sigma
# -0.616106 2.340852
The parameters for bull are mean = 1.703 and standard deviation = 1.483. The parameters for bear are mean = -0.616 and standard deviation = 2.341. They are close to the true values.
Bayseian Analysis
Here is an attempt to use Bayesian analysis to solve this question with jags and the rjags package.
I ran a Bayesian model to estimate the alpha (mean of y on bear), beta (The difference of y on bear and bull), and sigma (standard deviation of y on bear and bull) using 10000 iterations.
# Define the Bayesian model
model <- "model{
for(i in 1:length(Y)) {
Y[i] ~ dnorm(Mean[i], s[X[i]]^(-2))
Mean[i] <- alpha + beta[X[i]]
}
alpha ~ dnorm(0, 5^(-2))
beta[1] <- 0
beta[2] ~ dnorm(0, 5^(-2))
s[1] ~ dunif(0, 10)
s[2] ~ dunif(0, 10)
}"
# Compile the model
jags_model <- jags.model(
textConnection(model),
data = list(Y = dat$y, X = dat$regime),
n.chains = 3,
inits = list(.RNG.name = "base::Wichmann-Hill", .RNG.seed = 10)
)
# Simulate the posterior
jags_sim <- coda.samples(model = jags_model,
variable.names = c("alpha", "beta", "s"),
n.iter = 10000)
# Plot the posterior
plot(jags_sim)
The plot shows that estimates are well mixed.
# See the summary
summary(jags_sim)
Iterations = 1001:11000
Thinning interval = 1
Number of chains = 3
Sample size per chain = 10000
1. Empirical mean and standard deviation for each variable,
plus standard error of the mean:
Mean SD Naive SE Time-series SE
alpha -0.614 0.19436 0.0011222 0.0027201
beta[1] 0.000 0.00000 0.0000000 0.0000000
beta[2] 2.315 0.23099 0.0013336 0.0032666
s[1] 2.369 0.13768 0.0007949 0.0010393
s[2] 1.500 0.08836 0.0005102 0.0006727
2. Quantiles for each variable:
2.5% 25% 50% 75% 97.5%
alpha -0.9838 -0.7471 -0.6147 -0.4857 -0.2225
beta[1] 0.0000 0.0000 0.0000 0.0000 0.0000
beta[2] 1.8582 2.1622 2.3174 2.4760 2.7586
s[1] 2.1225 2.2722 2.3632 2.4564 2.6611
s[2] 1.3368 1.4390 1.4959 1.5579 1.6813
The mean of alpha on bear is -0.614, which is similar to the actual value -0.5. The mean of beta[2] is 2.315. If we add alpha and beta[2], we got 1.701, while the actual value is 1.7. We also got s[1] and s[2] as 2.369 and 1.5, which are similar to 2.3 and 1.6, respectively.
Bootstrapping
Here is another approch to use bootstrap to estimate alpha, beta, and standard deviation, which is based on the infer package.
# Load the infer package
library(infer)
set.seed(199)
# Split the data
dat_bull <- dat %>% filter(regime %in% "bull")
dat_bear <- dat %>% filter(regime %in% "bear")
# Calcualte the values in bull
dat_bull2 <- dat_bull %>%
# Specify the response variable
specify(response = y) %>%
# Generate 10000 bootstrap samples
generate(reps = 10000, type = "bootstrap")
summary_bull <- dat_bull2 %>%
summarise(mean_y = mean(y), sd_y = sd(y))
# Calcualte the values in bear
dat_ear2 <- dat_bear %>%
# Specify the response variable
specify(response = y) %>%
# Generate 10000 bootstrap samples
generate(reps = 10000, type = "bootstrap")
summary_bear <- dat_ear2 %>%
summarise(mean_y = mean(y), sd_y = sd(y))
Now we can print the results. The are all similar to the true values.
# The mean of bull
mean(summary_bull$mean_y)
# [1] 1.702693
# The standard deviation of bear
mean(summary_bull$sd_y)
# [1] 1.480158
# The mean of bear
mean(summary_bear$mean_y)
# [1] -0.6165585
# The standard deviation of bear
mean(summary_bear$sd_y)
# [1] 2.337042
In general: I want to calculate the (log) likelihood of data N given the estimated model parameters from data O.
More specifically, I want to know if my ll_given_modPars function below exists in one of the may R packages dealing with data modeling (lme4, glmm, etc.) as shown in this abstract example (not run):
library(lme4)
o_model <- lmer(observed ~ fixed.id + (1|random.id), data = O, REML = F)
n_logLik <- ll_given_modPars(model.estimates = o_model, data = N)
The fictional example above is on a linear mixed model for simplicity but I would like to eventually do this in a generalized linear mixed model which deals with the Poisson family or directly the negative binomial (for lme4: glmer(..., family="poisson") or glmer.nb ).
From what I could see most packages deal with parameter estimation (great, I need that) but then compare models on the same data with different combinations of fixed and random effects using anova or something to that extent which is not what I want to do.
I want the log likelihood for the same parameters on different data.
The main attempts made:
After not finding a function which seems to be doing that I thought of 'simply' tweaking the lme4 code to my purposes: it calculates the log likelihood for parameters given the data so I thought I could use the same framework but not have it optimize over different parameters but isolate the likelihood calculation function and just give it the parameters and the data. Unfortunately the code is a bit above my current skills https://github.com/lme4/lme4/blob/master/R/nbinom.R (I get a bit lost in how they use the objects over which they optimize).
I thought of doing the likelihood calculation myself, starting with a linear mixed model and then working my way up to more involved ones. But already with this example I'm having a hard time following the math and even when using the formula as specified the obtained log-likelihood is still different (I don't know why, see code in appendix) and I fear it will take me too long before I'll be able to do it for the more involved models (such as Poisson or negative binomial)
At this point I'm not sure what avenue is best to pursue and would appreciate any input you might have.
Appendix: Trying to calculate the log-likelihood (or finding a closed form approximation) based on How does lmer (from the R package lme4) compute log likelihood?. lmer (from lme4) gives a log-likelihood of -17.8 and I get -45.56
library(lme4)
set.seed(7)
n <- 2 # number of groups
m <- 4 # number of instances per group
fixed.effect <- c(0, -2, -1, 1)
tau <- 5 # standard deviation of random effects
sigma <- 2 # standard deviation of error
random.effect <- rnorm(n, mean=0, sd=tau)
sim.data <- data.frame(GROUP.ID=as.factor(rep(1:n, each=m)),
GROUP.EFFECT=rep(random.effect, each=m),
INSTANCE.ID=as.factor(rep(1:m, times=n)),
INSTANCE.EFFECT=rep(fixed.effect, times=n))
# calculate expected Y value
sim.data$EXPECT.Y <- sim.data$GROUP.EFFECT + sim.data$INSTANCE.EFFECT
# now observe Y value, assuming normally distributed with fixed std. deviation
sim.data$OBS.Y <- rnorm(nrow(sim.data), mean=sim.data$EXPECT.Y, sigma)
model <- lmer(OBS.Y ~ INSTANCE.ID + (1|GROUP.ID), data = sim.data, REML=F)
summary(model)
toy.model.var <- VarCorr(model)
toy.model.sigma <- attr(toy.model.var, 'sc') # corresponds to the epsilon, residual standard deviation
toy.model.tau.squared <- toy.model.var[[1]][1] # corresponds to variance of random effects
toy.model.betas <- model#beta
# left product, spread within gropus
toy.data <- rbind(sim.data$OBS.Y[1:4], sim.data$OBS.Y[5:8])
toy.mean.adj <- rbind(toy.data[1,] - mean(unlist(toy.data[1,])), toy.data[2,] - mean(unlist(toy.data[2,])))
toy.mean.adj.prod1 <- prod(dnorm(unlist(toy.mean.adj[1,]), mean = 0, sd = toy.model.sigma))
toy.mean.adj.prod2 <- prod(dnorm(unlist(toy.mean.adj[2,]), mean = 0, sd = toy.model.sigma))
toy.mean.adj.final.prod <- toy.mean.adj.prod1 * toy.mean.adj.prod2
# right product, spread between gropus
toy.mean.beta.adj <- rbind(mean(unlist(toy.data[1,])) - toy.model.betas, mean(unlist(toy.data[2,])) - toy.model.betas)
toy.mean.beta.adj[1,] <- toy.mean.beta.adj[1,] - c(0, toy.model.betas[1], toy.model.betas[1], toy.model.betas[1])
toy.mean.beta.adj[2,] <- toy.mean.beta.adj[2,] - c(0, toy.model.betas[1], toy.model.betas[1], toy.model.betas[1])
toy.mean.beta.adj.prod1 <- prod(dnorm(unlist(toy.mean.beta.adj[1,]), mean = 0, sd = sqrt(toy.model.sigma^2/4 + toy.model.tau.squared)) * sqrt(2/4*pi*toy.model.sigma^2))
toy.mean.beta.adj.prod2 <- prod(dnorm(unlist(toy.mean.beta.adj[2,]), mean = 0, sd = sqrt(toy.model.sigma^2/4 + toy.model.tau.squared)) * sqrt(2/4*pi*toy.model.sigma^2))
toy.mean.beta.adj.final.prod <- toy.mean.beta.adj.prod1 * toy.mean.beta.adj.prod2
toy.total.prod <- toy.mean.adj.final.prod * toy.mean.beta.adj.final.prod
log(toy.total.prod)
EDIT: A helpful link was provided in the comments (https://stats.stackexchange.com/questions/271903/understand-marginal-likelihood-of-mixed-effects-models). Converting my example from above I can replicate the log-likelihood
library(mvtnorm)
z = getME(model, "Z")
zt = getME(model, "Zt")
psi = bdiag(replicate(2, toy.model.tau.squared, simplify=FALSE))
betw = z%*%psi%*%zt
err = Diagonal(8, sigma(model)^2)
v = betw + err
dmvnorm(sim.data$OBS.Y, predict(model, re.form=NA), as.matrix(v), log=TRUE)
While I did not manage to come up with a closed form solution for all of them, I did manage to reproduce the log-likelihoods using numerical integration. I have posted below small examples for how it works in the LMM setting (assuming normal residuals random effects) as well as the GLMM with Poisson and Negative-Binomial. Note that especially the latter one tends so differ ever so slightly when you increase the sample size. My guess is that there is some rounding happening somewhere but for my purposes the precision achieved here is good enough. I will for now accept my own answer but if someone posts a closed form for the Poisson or the Negative-Binomial I will happily accept your answer :)
library(lme4)
library(mvtnorm)
################################################################################
# LMM numerical integration
set.seed(7)
n <- 2 # number of groups
m <- 4 # number of instances per group
fixed.effect <- c(0, -2, -1, 1)
tau <- 5 # standard deviation of random effects
sigma <- 2 # standard deviation of error
random.effect <- rnorm(n, mean=0, sd=tau)
normal.data <- data.frame(GROUP.ID=as.factor(rep(1:n, each=m)),
GROUP.EFFECT=rep(random.effect, each=m),
INSTANCE.ID=as.factor(rep(1:m, times=n)),
INSTANCE.EFFECT=rep(fixed.effect, times=n))
# calculate expected Y value
normal.data$EXPECT.Y <- normal.data$GROUP.EFFECT + normal.data$INSTANCE.EFFECT
# now observe Y value, assuming normally distributed with fixed std. deviation
normal.data$OBS.Y <- rnorm(nrow(normal.data), mean=normal.data$EXPECT.Y, sigma)
normal.model <- lmer(OBS.Y ~ INSTANCE.ID + (1|GROUP.ID), data = normal.data, REML=F)
summary(normal.model)
normal.model.var <- VarCorr(normal.model)
normal.model.sigma <- attr(normal.model.var, 'sc') # corresponds to the epsilon, residual standard deviation
normal.model.tau.squared <- normal.model.var[[1]][1] # corresponds to variance of random effects
normal.model.betas <- normal.model#beta
normal.group.tau <- sqrt(normal.model.tau.squared)
normal.group.sigma <- sigma(normal.model)
normal.group.beta <- predict(normal.model, re.form=NA)[1:4]
integrate_group1 <- function(x){
p1 <- dnorm(normal.data$OBS.Y[1] - normal.group.beta[1] - x, mean = 0, sd = normal.group.sigma) * dnorm(x, mean = 0, sd = normal.group.tau)
p2 <- dnorm(normal.data$OBS.Y[2] - normal.group.beta[2] - x, mean = 0, sd = normal.group.sigma)
p3 <- dnorm(normal.data$OBS.Y[3] - normal.group.beta[3] - x, mean = 0, sd = normal.group.sigma)
p4 <- dnorm(normal.data$OBS.Y[4] - normal.group.beta[4] - x, mean = 0, sd = normal.group.sigma)
p_out <- p1 * p2 * p3 * p4
p_out
}
normal.group1.integration <- integrate(integrate_group1, lower = -10*normal.group.tau, upper = 10*normal.group.tau, subdivisions = 10000L, rel.tol = 1e-10, abs.tol = 1e-50)$value[1]
integrate_group2 <- function(x){
p1 <- dnorm(normal.data$OBS.Y[5] - normal.group.beta[1] - x, mean = 0, sd = normal.group.sigma) * dnorm(x, mean = 0, sd = normal.group.tau)
p2 <- dnorm(normal.data$OBS.Y[6] - normal.group.beta[2] - x, mean = 0, sd = normal.group.sigma)
p3 <- dnorm(normal.data$OBS.Y[7] - normal.group.beta[3] - x, mean = 0, sd = normal.group.sigma)
p4 <- dnorm(normal.data$OBS.Y[8] - normal.group.beta[4] - x, mean = 0, sd = normal.group.sigma)
p_out <- p1 * p2 * p3 * p4
p_out
}
normal.group2.integration <- integrate(integrate_group2, lower = -10*normal.group.tau, upper = 10*normal.group.tau, subdivisions = 10000L, rel.tol = 1e-10, abs.tol = 1e-50)$value[1]
log(normal.group1.integration) + log(normal.group2.integration)
#################################
# Poisson numerical integration
set.seed(13) #13
n <- 2 # number of groups
m <- 4 # number of instances per group
# effect sizes are much smaller since they are exponentiated
fixed.effect <- c(0, -0.2, -0.1, 0.2)
tau <- 1.5 # standard deviation of random effects
# sigma <- 1.5 # standard deviation of error
random.effect <- rnorm(n, mean=0, sd=tau) # guide effect
poisson.data <- data.frame(GROUP.ID=as.factor(rep(1:n, each=m)),
GROUP.EFFECT=rep(random.effect, each=m),
INSTANCE.ID=as.factor(rep(1:m, times=n)),
INSTANCE.EFFECT=rep(fixed.effect, times=n))
# calculate expected Y value
poisson.data$EXPECT.Y <- exp(poisson.data$GROUP.EFFECT + poisson.data$INSTANCE.EFFECT)
# now observe Y value, assuming normally distributed with fixed std. deviation
poisson.data$OBS.Y <- rpois(nrow(poisson.data), poisson.data$EXPECT.Y)
poisson.model <- glmer(OBS.Y ~ INSTANCE.ID + (1|GROUP.ID), data = poisson.data, family="poisson")
summary(poisson.model)
poisson.model.var <- VarCorr(poisson.model)
poisson.model.sigma <- attr(poisson.model.var, 'sc') # corresponds to the epsilon, residual standard deviation
poisson.model.tau.squared <- poisson.model.var[[1]][1] # corresponds to variance of random effects
poisson.model.betas <- poisson.model#beta
poisson.group.tau <- sqrt(poisson.model.tau.squared)
poisson.group.sigma <- sigma(poisson.model)
poisson.group.beta <- predict(poisson.model, re.form=NA)[1:4]
integrate_group1 <- function(x){
p1 <- dpois(poisson.data$OBS.Y[1], lambda = exp(poisson.group.beta[1] + x)) * dnorm(x, mean = 0, sd = poisson.group.tau)
p2 <- dpois(poisson.data$OBS.Y[2], lambda = exp(poisson.group.beta[2] + x))
p3 <- dpois(poisson.data$OBS.Y[3], lambda = exp(poisson.group.beta[3] + x))
p4 <- dpois(poisson.data$OBS.Y[4], lambda = exp(poisson.group.beta[4] + x))
p_out <- p1 * p2 * p3 * p4
p_out
}
poisson.group1.integration <- integrate(integrate_group1, lower = -10*poisson.group.tau, upper = 10*poisson.group.tau, subdivisions = 10000L, rel.tol = 1e-10, abs.tol = 1e-50)$value[1]
integrate_group2 <- function(x){
p1 <- dpois(poisson.data$OBS.Y[5], lambda = exp(poisson.group.beta[1] + x)) * dnorm(x, mean = 0, sd = poisson.group.tau)
p2 <- dpois(poisson.data$OBS.Y[6], lambda = exp(poisson.group.beta[2] + x))
p3 <- dpois(poisson.data$OBS.Y[7], lambda = exp(poisson.group.beta[3] + x))
p4 <- dpois(poisson.data$OBS.Y[8], lambda = exp(poisson.group.beta[4] + x))
p_out <- p1 * p2 * p3 * p4
p_out
}
poisson.group2.integration <- integrate(integrate_group2, lower = -10*poisson.group.tau, upper = 10*poisson.group.tau, subdivisions = 10000L, rel.tol = 1e-10, abs.tol = 1e-50)$value[1]
log(poisson.group1.integration) + log(poisson.group2.integration)
#############
# Negative-Binomial numerical integration
set.seed(13) #13
n <- 100 # number of groups
m <- 4 # number of instances per group
# effect sizes are much smaller since they are exponentiated
fixed.effect <- c(0, -0.2, -0.1, 0.2)
tau <- 1.5 # standard deviation of random effects
theta <- 0.5
# sigma <- 1.5 # standard deviation of error
random.effect <- rnorm(n, mean=0, sd=tau) # guide effect
nb.data <- data.frame(GROUP.ID=as.factor(rep(1:n, each=m)),
GROUP.EFFECT=rep(random.effect, each=m),
INSTANCE.ID=as.factor(rep(1:m, times=n)),
INSTANCE.EFFECT=rep(fixed.effect, times=n))
# calculate expected Y value
nb.data$EXPECT.Y <- exp(nb.data$GROUP.EFFECT + nb.data$INSTANCE.EFFECT)
# now observe Y value, assuming normally distributed with fixed std. deviation
nb.data$OBS.Y <- rnbinom(nrow(nb.data), mu = nb.data$EXPECT.Y, size = theta)
nb.model <- glmer.nb(OBS.Y ~ INSTANCE.ID + (1|GROUP.ID), data = nb.data)
summary(nb.model)
nb.model.var <- VarCorr(nb.model)
nb.model.sigma <- attr(nb.model.var, 'sc') # corresponds to the epsilon, residual standard deviation
nb.model.tau.squared <- nb.model.var[[1]][1] # corresponds to variance of random effects
nb.model.betas <- nb.model#beta
nb.group.tau <- sqrt(nb.model.tau.squared)
nb.group.beta <- predict(nb.model, re.form=NA)[1:4]
nb.group.dispersion <- getME(nb.model, "glmer.nb.theta")
integration_function_generator <- function(input.obs, input.beta, input.dispersion, input.tau){
function(x){
p1 <- dnbinom(input.obs[1], mu = exp(input.beta[1] + x), size = input.dispersion) * dnorm(x, mean = 0, sd = input.tau)
p2 <- dnbinom(input.obs[2], mu = exp(input.beta[2] + x), size = input.dispersion)
p3 <- dnbinom(input.obs[3], mu = exp(input.beta[3] + x), size = input.dispersion)
p4 <- dnbinom(input.obs[4], mu = exp(input.beta[4] + x), size = input.dispersion)
p_out <- p1 * p2 * p3 * p4
p_out
}
}
nb.all.group.integrations <- c()
for(i in 1:n){
temp.obs <- nb.data$OBS.Y[(1:4)+(i-1)*4]
temp_integrate_function <- integration_function_generator(temp.obs, nb.group.beta, nb.group.dispersion, nb.group.tau)
temp.integration <- integrate(temp_integrate_function, lower = -10*nb.group.tau, upper = 10*nb.group.tau, subdivisions = 10000L, rel.tol = 1e-10, abs.tol = 1e-50)$value[1]
nb.all.group.integrations <- c(nb.all.group.integrations, temp.integration)
}
sum(log(nb.all.group.integrations))
I am a beginner of R so it may be a simple question.
I am now trying to fit a 4-dimensional point using thin-plate regression splines. One variable is a target variable and three variables are an explanatory variable.
I made a model with third order interaction and fitted the data to this.
library(mgcv)
dat <- read.csv('../data//data.csv')
model <- gam(Y ~ s(x1, x2, x3), data=dat)
By giving x3, I want to visualize a three-dimensional graph of spline curve or estimated contour plot, but how do I do it?
It will be very helpful if you can answer.
Thanks.
This is the sample data.
n = 100
x1 <- runif(n, min = 0, max = 100)
x2 <- runif(n, min = 0, max = 100)
x3 <- runif(n, min = 0, max = 100)
Y = numeric(n)
for(i in 1:n){
Y[i] <- x1[i]**0.5*x2[i]**2*x3[i]/10000
}
dat = data.frame(Y=Y, x1=x1, x2=x2, x3=x3)
I do thin-plane regression spline using this dat.
model <- gam(Y ~ s(x1, x2, x3, k= 50), data=dat)
Then, I would like to obtain a fitting curve of three-dimensional thin-plane regression spline or contour plot estimated by regression spline when x3 = 25, for example.
To make a contour plot, you can use contour(x, y, z, ...). z is your data Matrix (in your case, Y[x1,x2, ], x and y are index vectors from 0 to 1 with a length of nrow(Y[x1,x2, ]) and ncol(Y[x1,x2, ]).
You should be able to use it similar to:
contour( x = seq(0, 1, length.out = length(x1)), y = seq(0, 1, length.out = length(x2)), z = Y[x1,x2, ] )
I found a solution with reference to the answer of d0d0.
n=100
const=25
x = y = seq(0, n, 1)
f = function(x,y){
dtmp <- data.frame(x1=(x), x2=(y), x3=(const))
pred <- predict.gam(model, dtmp)
}
z = outer(x, y, f)
contour(x,y,z)
Say I have 200 subjects, 100 in group A and 100 in group B, and for each I measure some continuous parameter.
require(ggplot2)
set.seed(100)
value <- c(rnorm(100, mean = 5, sd = 3), rnorm(100, mean = 10, sd = 3))
group <- c(rep('A', 100), rep('B', 100))
data <- data.frame(value, group)
ggplot(data = data, aes(x = value)) +
geom_bar(aes(color = group))
I would like to determine the value (Threshold? Breakpoint?) that maximizes separation and minimizes misclassification between the groups. Does such a function exist in R?
I've tried searching along the lines of "r breakpoint maximal separation between groups," and "r threshold minimize misclassification," but my google-foo seems to be off today.
EDIT:
Responding to #Thomas's comment, I have tried to fit the data using logistic regression and then solve for the threshold, but I haven't gotten very far.
lr <- glm(group~value)
coef(lr)
# (Intercept) value
# 1.1857435 -0.0911762
So Bo = 1.1857435 and B1 = -0.0911762
From Wikipedia, I see that F(x) = 1/(1+e^-(Bo + B1x)), and solving for x:
x = (ln(F(x) / (1 - F(x))) - Bo)/B1
But trying this in R, I get an obviously incorrect answer:
(log(0.5/(1 - 0.5)) - 1.1857435)/-0.0911762 # 13.00497
A simple approach is to write a function that calculates the accuracy given a threshold:
accuracy = Vectorize(function(th) mean(c("A", "B")[(value > th) + 1] == group))
Then find the maximum using optimize:
optimize(accuracy, c(min(value), max(value)), maximum=TRUE)
# $maximum
# [1] 8.050888
#
# $objective
# [1] 0.86
I've gotten the answer I need thanks to help from #Thomas and #BenBolker.
Summary
The problem with my attempt at solving it through logistic regression was that I hadn't specified family = binomial
The dose.p() function in MASS will do the work for me given a glm fit
Code
# Include libraries
require(ggplot2)
require(MASS)
# Set seed
set.seed(100)
# Put together some dummy data
value <- c(rnorm(100, mean = 5, sd = 3), rnorm(100, mean = 10, sd = 3))
group <- c(rep(0, 100), rep(1, 100))
data <- data.frame(value, group)
# Plot the distribution -- visually
# The answer appears to be b/t 7 and 8
ggplot(data = data, aes(x = value)) +
geom_bar(aes(color = group))
# Fit a glm model, specifying the binomial distribution
my.glm <- glm(group~value, data = data, family = binomial)
b0 <- coef(my.glm)[[1]]
b1 <- coef(my.glm)[[2]]
# See what the probability function looks like
lr <- function(x, b0, b1) {
prob <- 1 / (1 + exp(-1*(b0 + b1*x)))
return(prob)
}
# The line appears to cross 0.5 just above 7.5
x <- -0:12
y <- lr(x, b0, b1)
lr.val <- data.frame(x, y)
ggplot(lr.val, aes(x = x, y = y)) +
geom_line()
# The inverse of this function computes the threshold for a given probability
inv.lr <- function(p, b0, b1) {
x <- (log(p / (1 - p)) - b0)/b1
return(x)
}
# With the betas from this function, we get 7.686814
inv.lr(0.5, b0, b1)
# Or, feeding the glm model into dose.p from MASS, we get the same answer
dose.p(my.glm, p = 0.5)
Thanks, everyone, for your help!