I have a study with several cases, all containing data from multiple ordinal factor variables (genotypes) and multiple numeric variables (various blood samples (concentrations)). I am trying to set up an explorative model to test linearity between any of the numeric variables (dependent in the model) and any of the ordinal factor variables (independent in the model).
Dataset structure example (independent variables): genotypes
case_id genotype_1 genotype_2 ... genotype_n
1 0 0 1
2 1 0 2
... ... ... ...
n 2 1 0
and dependent variables (with matching case id:s): samples
case_id sample_1 sample_2 ... sample_n
1 0.3 0.12 6.12
2 0.25 0.15 5.66
... ... ... ...
n 0.44 0.26 6.62
Found one similar example in the forum which doesn't solve the problem:
model <- apply(samples,2,function(xl)lm(xl ~.,data= genotypes))
I can't figure out how to make simple linear regressions that go through any combination of a given set of dependent and independent variables. If using apply family I guess the varying (x) term should be the dependent variable in the model since every dependent variable should test linearity for the same set of independent variables (individually).
Extract from true data:
> genotypes
case_id genotype_1 genotype_2 genotype_3 genotype_4 genotype_5
1 1 2 2 1 1 0
2 2 NaN 1 NaN 0 0
3 3 1 0 0 0 NaN
4 4 2 2 1 1 0
5 5 0 0 0 1 NaN
6 6 2 2 1 0 0
7 9 0 0 0 0 1
8 10 0 0 0 NaN 0
9 13 0 0 0 NaN 0
10 15 NaN 1 NaN 0 1
> samples
case_id sample_1 sample_2 sample_3 sample_4 sample_5
1 1 0.16092019 0.08814160 -0.087733372 0.1966070 0.09085343
2 2 -0.21089678 -0.13289427 0.056583528 -0.9077926 -0.27928376
3 3 0.05102400 0.07724300 -0.212567535 0.2485348 0.52406368
4 4 0.04823619 0.12697286 0.010063683 0.2265085 -0.20257192
5 5 -0.04841221 -0.10780329 0.005759269 -0.4092782 0.06212171
6 6 -0.08926734 -0.19925538 0.202887833 -0.1536070 -0.05889369
7 9 -0.03652588 -0.18442457 0.204140717 0.1176950 -0.65290133
8 10 0.07038933 0.05797007 0.082702589 0.2927817 0.01149564
9 13 -0.14082554 0.26783539 -0.316528107 -0.7226103 -0.16165326
10 15 -0.16650266 -0.35291579 0.010063683 0.5210507 0.04404433
SUMMARY: Since I have a lot of data I want to create a simple model to help me select which possible correlations to look further into. Any ideas out there?
NOTE: I am not trying to fit a multiple linear regression model!
I feel like there must be a statistical test for linearity, but I can't recall it. Visual inspection is typically how I do it. Quick and dirty way to test for linearity for a large number of variables would be to test the corr() of each pair of dependent/independent variables. Small multiples would be a handy way to do it.
Alternately, for each dependent ordinal variable, run a corrplot vs. each independent (numerical) variable, a logged version of the independent variable, and the exponentiated version of the independant variable. If the result of CORR for the logged or exponented version has a higher p-value than the regular version, it seems likely you have some linearity issues.
Related
I have a dataset with four variables (df)
household
group
income
post
1
0
20'000
0
1
0
22'000
1
2
1
10'000
0
2
1
20'000
1
3
0
20'000
0
3
0
21'000
1
4
1
9'000
0
4
1
16'000
1
5
1
8'000
0
5
1
18'000
1
6
0
22'000
0
6
0
26'000
1
7
1
12'000
0
7
1
24'000
1
8
0
24'000
0
8
0
27'000
1
Group is a binary variable and is 1, when household got support from state. and post variable is also binary and is 1, when it is after some household got support from state.
Now I would like to run a before vs after regression that estimates the group effect by comparing post-period and before period for the supported group. I would like to put the dependent variable in logs, to have the effect in percentage, so the impact of state support on income.
I used that code, but I don't know if it is right to get the answer?
library("fixest")
feols(log(income) ~ group + post,data=df) %>% etable()
Is there another way?
If you are looking for the classic 2x2 design your code was almost correct. Change '+' with '*'. This tell us that the supported group increased the income with 7 250 more than the group which not received support.
comparing = feols(income ~ group * post,data)
comparing_log = feols(log(income) ~ group * post,data)
etable(comparing,comparing_log)
PS: The interpretation of the coefficient as percentage change is a good approximation for small numbers. The correct formula for % change is: exp(beta)-1. In this case it is exp(0.5829)-1 = 0.7912.
So the change here is 79,12%.
Is it possible to split episode by a given variable in survival analysis in R, similar to in STATA using stsplit in the following way: stsplit var, at(0) after(time=time)?
I am aware that the survival package allows one to split episode by given cut points such as c(0,5,10,15) in survSplit, but if a variable, say time of divorce, differs by each individual, then providing cutpoints for each individual would be impossible, and the split would have to be based on the value of a variable (say graduation, or divorce, or job termination).
Is anyone aware of a package or know a resource I might be able to tap into?
Perhaps Epi package is what you are looking for. It offers multiple ways to cut/split the follow-up time using the Lesix objects. Here is the documentation of cutLesix().
After some poking around, I think tmerge() in the survival package can achieve what stsplit var can do, which is to split episodes not just by a given cut points (same for all observations), but by when an event occurs for an individual.
This is the only way I knew how to split data
id<-c(1,2,3)
age<-c(19,20,29)
job<-c(1,1,0)
time<-age-16 ## create time since age 16 ##
data<-data.frame(id,age,job,time)
id age job time
1 1 19 1 3
2 2 20 1 4
3 3 29 0 13
## simple split by time ##
## 0 to up 2 years, 2-5 years, 5+ years ##
data2<-survSplit(data,cut=c(0,2,5),end="time",start="start",
event="job")
id age start time job
1 1 19 0 2 0
2 1 19 2 3 1
3 2 20 0 2 0
4 2 20 2 4 1
5 3 29 0 2 0
6 3 29 2 5 0
7 3 29 5 13 0
However, if I want to split by a certain variable, such as when each individuals finished school, each person might have a different cut point (finished school at different ages).
## split by time dependent variable (age finished school) ##
d1<-data.frame(id,age,time,job)
scend<-c(17,21,24)-16
d2<-data.frame(id,scend)
## create start/stop time ##
base<-tmerge(d1,d1,id=id,tstop=time)
## create time-dependent covariate ##
s1<-tmerge(base,d2,id=id,
finish=tdc(scend))
id age time job tstart tstop finish
1 1 19 3 1 0 1 0
2 1 19 3 1 1 3 1
3 2 20 4 1 0 4 0
4 3 29 13 0 0 8 0
5 3 29 13 0 8 13 1
I think tmerge() is more or less comparable with stsplit function in STATA.
I am trying to compute robust/cluster standard errors after using mlogit() to fit a Multinomial Logit (MNL) in a Discrete Choice problem. Unfortunately, I suspect I am having problems with it because I am using data in long format (this is a must in my case), and getting the error #Error in ef/X : non-conformable arrays after sandwich::vcovHC( , "HC0").
The Data
For illustration, please gently consider the following data. It represents data from 5 individuals (id_ind ) that choose among 3 alternatives (altern). Each of the five individuals chose three times; hence we have 15 choice situations (id_choice). Each alternative is represented by two generic attributes (x1 and x2), and the choices are registered in y (1 if selected, 0 otherwise).
df <- read.table(header = TRUE, text = "
id_ind id_choice altern x1 x2 y
1 1 1 1 1.586788801 0.11887832 1
2 1 1 2 -0.937965347 1.15742493 0
3 1 1 3 -0.511504401 -1.90667519 0
4 1 2 1 1.079365680 -0.37267925 0
5 1 2 2 -0.009203032 1.65150370 1
6 1 2 3 0.870474033 -0.82558651 0
7 1 3 1 -0.638604013 -0.09459502 0
8 1 3 2 -0.071679538 1.56879334 0
9 1 3 3 0.398263302 1.45735788 1
10 2 4 1 0.291413453 -0.09107974 0
11 2 4 2 1.632831160 0.92925495 0
12 2 4 3 -1.193272276 0.77092623 1
13 2 5 1 1.967624379 -0.16373709 1
14 2 5 2 -0.479859282 -0.67042130 0
15 2 5 3 1.109780885 0.60348187 0
16 2 6 1 -0.025834772 -0.44004183 0
17 2 6 2 -1.255129594 1.10928280 0
18 2 6 3 1.309493274 1.84247199 1
19 3 7 1 1.593558740 -0.08952151 0
20 3 7 2 1.778701074 1.44483791 1
21 3 7 3 0.643191170 -0.24761157 0
22 3 8 1 1.738820924 -0.96793288 0
23 3 8 2 -1.151429915 -0.08581901 0
24 3 8 3 0.606695064 1.06524268 1
25 3 9 1 0.673866953 -0.26136206 0
26 3 9 2 1.176959443 0.85005871 1
27 3 9 3 -1.568225496 -0.40002252 0
28 4 10 1 0.516456176 -1.02081089 1
29 4 10 2 -1.752854918 -1.71728381 0
30 4 10 3 -1.176101700 -1.60213536 0
31 4 11 1 -1.497779616 -1.66301234 0
32 4 11 2 -0.931117325 1.50128532 1
33 4 11 3 -0.455543630 -0.64370825 0
34 4 12 1 0.894843784 -0.69859139 0
35 4 12 2 -0.354902281 1.02834859 0
36 4 12 3 1.283785176 -1.18923098 1
37 5 13 1 -1.293772990 -0.73491317 0
38 5 13 2 0.748091387 0.07453705 1
39 5 13 3 -0.463585127 0.64802031 0
40 5 14 1 -1.946438667 1.35776140 0
41 5 14 2 -0.470448172 -0.61326604 1
42 5 14 3 1.478763383 -0.66490028 0
43 5 15 1 0.588240775 0.84448489 1
44 5 15 2 1.131731049 -1.51323232 0
45 5 15 3 0.212145247 -1.01804594 0
")
The problem
Consequently, we can fit an MNL using mlogit() and extract their robust variance-covariance as follows:
library(mlogit)
library(sandwich)
mo <- mlogit(formula = y ~ x1 + x2|0 ,
method ="nr",
data = df,
idx = c("id_choice", "altern"))
sandwich::vcovHC(mo, "HC0")
#Error in ef/X : non-conformable arrays
As we can see there is an error produced by sandwich::vcovHC, which says that ef/X is non-conformable. Where X <- model.matrix(x) and ef <- estfun(x, ...). After looking through the source code on the mirror on GitHub I spot the problem which comes from the fact that, given that the data is in long format, ef has dimensions 15 x 2 and X has 45 x 2.
My workaround
Given that the show must continue, I am computing the robust and cluster standard errors manually using some functions that I borrow from sandwich and I adjusted to accommodate the Stata's output.
> Robust Standard Errors
These lines are inspired on the sandwich::meat() function.
psi<- estfun(mo)
k <- NCOL(psi)
n <- NROW(psi)
rval <- (n/(n-1))* crossprod(as.matrix(psi))
vcov(mo) %*% rval %*% vcov(mo)
# x1 x2
# x1 0.23050261 0.09840356
# x2 0.09840356 0.12765662
Stata Equivalent
qui clogit y x1 x2 ,group(id_choice) r
mat li e(V)
symmetric e(V)[2,2]
y: y:
x1 x2
y:x1 .23050262
y:x2 .09840356 .12765662
> Clustered Standard Errors
Here, given that each individual answers 3 questions is highly likely that there is some degree of correlation among individuals; hence cluster corrections should be preferred in such situations. Below I compute the cluster correction in this case and I show the equivalence with the Stata output of clogit , cluster().
id_ind_collapsed <- df$id_ind[!duplicated(mo$model$idx$id_choice,)]
psi_2 <- rowsum(psi, group = id_ind_collapsed )
k_cluster <- NCOL(psi_2)
n_cluster <- NROW(psi_2)
rval_cluster <- (n_cluster/(n_cluster-1))* crossprod(as.matrix(psi_2))
vcov(mo) %*% rval_cluster %*% vcov(mo)
# x1 x2
# x1 0.1766707 0.1007703
# x2 0.1007703 0.1180004
Stata equivalent
qui clogit y x1 x2 ,group(id_choice) cluster(id_ind)
symmetric e(V)[2,2]
y: y:
x1 x2
y:x1 .17667075
y:x2 .1007703 .11800038
The Question:
I would like to accommodate my computations within the sandwich ecosystem, meaning not computing the matrices manually but actually using the sandwich functions. Is it possible to make it work with models in long format like the one described here? For example, providing the meat and bread objects directly to perform the computations? Thanks in advance.
PS: I noted that there is a dedicated bread function in sandwich for mlogit, but I could not spot something like meat for mlogit, but anyways I am probably missing something here...
Why vcovHC does not work for mlogit
The class of HC covariance estimators can just be applied in models with a single linear predictor where the score function aka estimating function is the product of so-called "working residuals" and a regressor matrix. This is explained in some detail in the Zeileis (2006) paper (see Equation 7), provided as vignette("sandwich-OOP", package = "sandwich") in the package. The ?vcovHC also pointed to this but did not explain it very well. I have improved this in the documentation at http://sandwich.R-Forge.R-project.org/reference/vcovHC.html now:
The function meatHC is the real work horse for estimating the meat of HC sandwich estimators - the default vcovHC method is a wrapper calling sandwich and bread. See Zeileis (2006) for more implementation details. The theoretical background, exemplified for the linear regression model, is described below and in Zeileis (2004). Analogous formulas are employed for other types of models, provided that they depend on a single linear predictor and the estimating functions can be represented as a product of “working residual” and regressor vector (Zeileis 2006, Equation 7).
This means that vcovHC() is not applicable to multinomial logit models as they generally use separate linear predictors for the separate response categories. Similarly, two-part or hurdle models etc. are not supported.
Basic "robust" sandwich covariance
Generally, for computing the basic Eicker-Huber-White sandwich covariance matrix estimator, the best strategy is to use the sandwich() function and not the vcovHC() function. The former works for any model with estfun() and bread() methods.
For linear models sandwich(..., adjust = FALSE) (default) and sandwich(..., adjust = TRUE) correspond to HC0 and HC1, respectively. In a model with n observations and k regression coefficients the former standardizes with 1/n and the latter with 1/(n-k).
Stata, however, divides by 1/(n-1) in logit models, see:
Different Robust Standard Errors of Logit Regression in Stata and R. To the best of my knowledge there is no clear theoretical reason for using specifically one or the other adjustment. And already in moderately large samples, this makes no difference anyway.
Remark: The adjustment with 1/(n-1) is not directly available in sandwich() as an option. However, coincidentally, it is the default in vcovCL() without specifying a cluster variable (i.e., treating each observation as a separate cluster). So this is a convenient "trick" if you want to get exactly the same results as Stata.
Clustered covariance
This can be computed "as usual" via vcovCL(..., cluster = ...). For mlogit models you just have to consider that the cluster variable just needs to be provided once (as opposed to stacked several times in long format).
Replicating Stata results
With the data and model from your post:
vcovCL(mo)
## x1 x2
## x1 0.23050261 0.09840356
## x2 0.09840356 0.12765662
vcovCL(mo, cluster = df$id_choice[1:15])
## x1 x2
## x1 0.1766707 0.1007703
## x2 0.1007703 0.1180004
I am trying to train a model using bstTree method and print out the confusion matrix. adverse_effects is my class attribute.
set.seed(1234)
splitIndex <- createDataPartition(attended_num_new_bstTree$adverse_effects, p = .80, list = FALSE, times = 1)
trainSplit <- attended_num_new_bstTree[ splitIndex,]
testSplit <- attended_num_new_bstTree[-splitIndex,]
ctrl <- trainControl(method = "cv", number = 5)
model_bstTree <- train(adverse_effects ~ ., data = trainSplit, method = "bstTree", trControl = ctrl)
predictors <- names(trainSplit)[names(trainSplit) != 'adverse_effects']
pred_bstTree <- predict(model_bstTree$finalModel, testSplit[,predictors])
plot.roc(auc_bstTree)
conf_bstTree= confusionMatrix(pred_bstTree,testSplit$adverse_effects)
But I get the error 'Error in confusionMatrix.default(pred_bstTree, testSplit$adverse_effects) :
The data must contain some levels that overlap the reference.'
max(pred_bstTree)
[1] 1.03385
min(pred_bstTree)
[1] 1.011738
> unique(trainSplit$adverse_effects)
[1] 0 1
Levels: 0 1
How can I fix this issue?
> head(trainSplit)
type New_missed Therapytypename New_Diesease gender adverse_effects change_in_exposure other_reasons other_medication
5 2 1 14 13 2 0 0 0 0
7 2 0 14 13 2 0 0 0 0
8 2 0 14 13 2 0 0 0 0
9 2 0 14 13 2 1 0 0 0
11 2 1 14 13 2 0 0 0 0
12 2 0 14 13 2 0 0 0 0
uvb_puva_type missed_prev_dose skintypeA skintypeB Age DoseB DoseA
5 5 1 1 1 22 3.000 0
7 5 0 1 1 22 4.320 0
8 5 0 1 1 22 4.752 0
9 5 0 1 1 22 5.000 0
11 5 1 1 1 22 5.000 0
12 5 0 1 1 22 5.000 0
I had similar problem, which refers to this error. I used function confusionMatrix:
confusionMatrix(actual, predicted, cutoff = 0.5)
An I got the following error: Error in confusionMatrix.default(actual, predicted, cutoff = 0.5) : The data must contain some levels that overlap the reference.
I checked couple of things like:
class(actual) -> numeric
class(predicted) -> integer
unique(actual) -> plenty values, since it is probability
unique(predicted) -> 2 levels: 0 and 1
I concluded, that there is problem with applying cutoff part of the function, so I did it before by:
predicted<-ifelse(predicted> 0.5,1,0)
and run the confusionMatrix function, which works now just fine:
cm<- confusionMatrix(actual, predicted)
cm$table
which generated correct outcome.
One takeaway for your case, which might improve interpretation once you make code working:
you mixed input values for your confusion matrix(as per confusionMatrix package documetation), instead of:
conf_bstTree= confusionMatrix(pred_bstTree,testSplit$adverse_effects)
you should have written:
conf_bstTree= confusionMatrix(testSplit$adverse_effects,pred_bstTree)
As said it will most likely help you interpret confusion matrix, once you figure out way to make it work.
Hope it helps.
max(pred_bstTree) [1] 1.03385
min(pred_bstTree) [1] 1.011738
and errors tells it all. Plotting ROC is simply checking the effect of different threshold points. Based on threshold rounding happens e.g. 0.7 will be converted to 1 (TRUE class) and 0.3 will be go 0 (FALSE class); in case threshold is 0.5. Threshold values are in range of (0,1)
In your case regardless of threshold you will always get all observations into TRUE class as even minimum prediction is greater than 1. (Thats why #phiver was wondering if you are doing regression instead of classification) . Without any zero in prediction there is no level in 'prediction' which coincide with zero level in adverse_effects and hence this error.
PS: It will be difficult to tell root cause of error without you posting your data
I am trying to use svm() to classify my data. A sample of my data is as follows:
ID call_YearWeek week WeekCount oc
x 2011W01 1 0 0
x 2011W02 2 1 1
x 2011W03 3 0 0
x 2011W04 4 0 0
x 2011W05 5 1 1
x 2011W06 6 0 0
x 2011W07 7 0 0
x 2011W08 8 1 1
x 2011W09 9 0 0
x 2011W10 10 0 0
x 2011W11 11 0 0
x 2011W12 12 1 1
x 2011W13 13 1 1
x 2011W14 14 1 1
x 2011W15 15 0 0
x 2011W16 16 2 1
x 2011W17 17 0 0
x 2011W18 18 0 0
x 2011W19 19 1 1
The third column shows week of the year. The 4th column shows number of calls in that week and the last column is a binary factor (if a call was received in that week or not). I used the following lines of code:
train <- data[1:105,]
test <- data[106:157,]
model <- svm(oc~week,data=train)
plot(model,train,week)
plot(model,train)
none of the last two lines work. they dont show any plots and they return no error. I wonder why this is happening.
Thanks
Seems like there are two problems here, first is that not all svm types are supported by plot.svm -- only the classification methods are, and not the regression methods. Because your response is numeric, svm() assumes you want to do regression so it chooses "eps-regression" by default. If you want to do classification, change your response to a factor
model <- svm(factor(oc)~week,data=train)
which will then use "C-classification" by default.
The second problem is that there does not seem to be a univariate predictor plot implemented. It seems to want two variables (one for x and one for y).
It may be better to take a step back and describe exactly what you want your plot to look like.