YawMMF

Effective Mixed Model Functions

Citation

Zhang, G., Li, X., & Lin, N. (2020). YawMMF: Effective Mixed Model Functions. Available at: https://github.com/usplos/YawMMF.

Install

if(!require(devtools)) install.packages("devtools")
devtools::install_github("usplos/YawMMF")

If not necessary, do not update the packages suggested to be updated.

If not necessary, do not update the packages suggested to be updated.

If not necessary, do not update the packages suggested to be updated.

Installation Bugs and Possible Solutions

Bug 01

> install.packages("YawMMF")
Warning in install.packages :
  package ‘YawMMF’ is not available (for R version 3.6.1)

YawMMF package should be installed from Github instead of CRAN

Solution

Please use the command devtools::install_github('usplos/YawMMF') instead of install.packages("YawMMF")

Bug 02

WARNING: Rtools is required to build R packages, but is not currently installed.

Please download and install Rtools 3.5 from http://cran.r-project.org/bin/windows/Rtools/

Solution 1

Download and install Rtools.

Solution 2

Download and install Git.

Bug 03

* installing *source* package 'YawMMF' ...
** using staged installation
** R
** data
*** moving datasets to lazyload DB
** byte-compile and prepare package for lazy loading
错误: (由警告转换成)程辑包'simr'是用R版本3.6.2 来建造的
停止执行
ERROR: lazy loading failed for package 'YawMMF'
Error: Failed to install 'YawMMF' from GitHub:
  (converted from warning) installation of package ‘YawMMF’ had non-zero exit status

Solution

Update R to the latest version, because the latest versions of some packages (e.g., simr) also require the latest version of R.

Tips: You can use the installr package to copy all your installed packages from the old folder to the new one.

install.packages("installr")
library(installr)
copy.packages.between.libraries(ask=TRUE)

Brief introduction

This package offers convenient and effective functions for mixed model.

• EffectSize() : to calculate the effect size for mixed model;

• MixedModelDataExplore() : to explore the dependent vairable (usually a continuous variable)

• MixedModelDataSummary() : to summarise your data. (This function is still maturing)

• MixedModelDummy() : to generate dummy variables for factor variables.

• MixedModelDiag() : to offer advice so as to optimize your model.

• MixedModelPower() : to perform the power analysis.

• MixedModelPlot() : to draw bar plot, interaction line plot, violin plus box plot, violin plus rawdata plot.

• MixedModelWrite() : to write the fixed effects of the given model or the descriptive inforamtion of the given dependent variable into yourdisk (only works only works for one model at one time)

• MixedModelWrite2(): to write the fixed effects of the given model or the descriptive inforamtion of the given dependent variable into yourdisk (works for multiple models at one time)

• CorrPlot() : to draw correlation-matrix plot for multiple-variable correlations

• contr.simple() : to set the contrast matrix of factor variables as simple contrast.

Update log

20210119

A new fucntion named matchingvariablegeneralized(y, by, type = 'F', threshold = 1) has been updated. matchingvariable() can find as many obsearved data as possible that are not significantly different between only two conditions (or within one condition variable with two levels). However, sometimes there are more than two levels within one condition variable. In addition, usually there are more than one condition variable. In these situation, matchingvariable() can not satify the users, while matchingvariablegeneralized() can.

When using this function, users should set the parameter by as a list object which include the condition variables of interest.

set.seed(1234)
y = c(rnorm(100,0),
      rnorm(120,0.1),
      rnorm(120,0.2),
      rnorm(110,0.3),
      rnorm(110,0.2),
      rnorm(120,0.4))
by = cbind(c(rep('A1',340),rep('A2',340)),
           c(rep('B1',100),rep('B2',120),rep('B3',120),
             rep('B1',110), rep('B2',110), rep('B3',120)))
testdf = tibble(y,V1=by[,1],V2=by[,2])

testdf = testdf %>% 
  mutate(matchingtest = matchingvariablegeneralized(y = y, 
                                                    by = list(V1,V2), 
                                                    type = 'F',
                                                    threshold = 1.5)) %>%
  filter(matchingtest == 1)

20210117

A new function named matchingvalue(y, value, type='t', threshold = 1) has been updated. This function works in a similar way with matchingvariable(). It can find as many obsearved data (set using the parameter y in the function) as possible of which the mean is not significantly different with a given value (set using the parameter value in the function).

set.seed(1234)
y = rnorm(n = 200, mean = 0.5)

matchingvalue(y = y, value, type='t', threshold = 1)

20210107

A new function named matchingvariable(y = , by = , type = 't', threshold = 1) has been updated. It is very common in cognitive experiments that you need two sets of experimental materials that are equal in number and match on a particular attribute. You want to make sure that there is no significant difference between the two groups of materials in the (rating) score for this attribute, and you want to find as many materials as possible that meet this requirement. This function will help you do this efficiently.

This function will return a set of 0-1 vector, in which 1 means qualified, while 0 means unqualified.

You can use statistical t value (e.g. t < 1) to select the materials:

y1 = rnorm(100,mean = 0, sd = 1)
cond1 = rep('C1', 100)
y2 = rnorm(120,mean = 0.5, sd = 1)
cond2 = rep('C2', 120)
y = c(y1, y2)
by = c(cond1, cond2)
matchingvariable(y = y, by = by, type = 't', threshold = 1)

You can also use statistical p value (e.g. p > 0.1) to select the materials:

matchingvariable(y = y, by = by, type = 'p', threshold = 0.1)

20201217

A new function named PCOR(X,Y,...,Method = 'pearson', Alternative='two.sided') has been updated. It can perfrom the partial correlation analysis. The X and 'Y' are the two variables that the partial correlation is perfomed between. ... are the variables to be controlled. Method and Alternative parameters can be defined in the same way as the method and alternative ones in the base R function cor.test().

20201028

Critical update: A new function named MixedModelWrite2(Model = NULL, Data = NULL, Prefix = 'DV', compareModel = F, ModelnamesTocompare) has been created. It can output the information of fixed effects for more than one model in the form of docx document. This function requires export package and texreg package. But these can be automatically installed.

The information of parameters is following:

• Model - a list of Models;
• Data - a list of summarised data;
• Prefix - prefix for the name of the output file;
• compareModel - whether to compare different models; it only works when the Model is not set NULL;
• ModelnamesTocompare - the names of the models to be compared; it only works when the compareModel is set 'T'.

Here is one example:

library(YawMMF)
Model1 = lmer(data = DemoData, 
              DV~CondA+(1|subj)+(1|item))
Model2 = lmer(data = DemoData, 
              log(DV)~CondA+(1|subj)+(1|item))
MixedModelWrite2(Model = list(Model1, Model2),
                 Prefix = 'DV',
                 compareModel = T,
                 ModelnamesTocompare = c('Origin','Log'))

20201016

• A new function named EffectSize() has been added. This function depends on effectsize package in R, and can be used to calculate the effect size of the fixed effects for a given mixed model (whether the model is generalized or not).

Notion that the parameter Type can only be defined as 'd' or 'r'.

Notion that the function will provide the equation of calculating accordingly, which will help researchers understand and check the results.

Here is an example:

library(YawMMF)
Model1 = lmer(data=DemoData, DV~CondA+(1|subj))
Model2 = glmer(data = DemoData2, DV~CondA+(1|subj),family = 'binomial')

# to calculate the cohen d value
EffectSize(Model = Model1, Type = 'd', GLMM = F)
EffectSize(Model = Model2, Type = 'd', GLMM = T)

# to calculate the r value
EffectSize(Model = Model1, Type = 'r', GLMM = F)
EffectSize(Model = Model2, Type = 'r', GLMM = T)

20200918

• A new funciton named StrInsertSpace() has been added. Using it like StrInsertSpace(c('AA','AAA','AAAA'))
• A new function named StrDeletePosition() has been added. Using it like StrDeletePosition(X = 'ABCDEFG', position = c(1,3,5)
• A new funciton named StrSubPosition() has been added. Using it like StrSubPosition(X = 'ABCDEFG', position = c(1,3,5)
• A new function named StrUniqueChar() has been added. Using it like StrUniqueChar(X = 'ABCDDEFGG')

20200831

• A new function named p.sig(p, numcontrasts = 1) has been updated. This function can return the significance labels for p value. For the parameter numcontrasts, the default value is 1, indicating that there is no Bonferroni correction. And if any, by setting this parameter, this function can return the significance labels whether the contrasts have survived the Bonferroni correction.

20200621

• The function MixedModelOpt() has been removed. There are some potential problems within it, which may cause misleading results.
• A new function named MixedModelDiag() has been updated. Unlike MixedModelOpt(), this new function does not return the optimized model and corresponse results, but provides more comprehensive and objective information which the users can refer to. Here is an example for using it.
• Several convenient functions have been updated. Simplecoding() is able to set the contrast-coding for factors within the data. rePCA2df() can help tidy the results of rePCA(). VarCorr2df() can help tidy the results of VarCorr().

library(YawMMF)
DemoData %>% 
    Simplecoding(data = ., Factor = 'CondA,CondB') %>%
    MixedModelDummy(data = ., Fix_Factor = 'CondA,CondB',MatrixDesign = '*',ContrastsM = F) %>% 
    MixedModelDiag(data = ., 
                   DV = 'DV', 
                   IV = 'CondA*CondB',
                   randomfactor = 'subj,item',
                   randomeffect = '1+CondA1+CondB1+CondA1_CondB1,1+CondA1+CondB1+CondA1_CondB1',
                   PCAdeletecriterion = 0)

20200415

• A new function named NORM_CLUST() has been added. This function can seperate a given sample, of which distribution is not normal, into a limited number of sub-sample of which distributions are all normal, and then return the numeric category labels.

20200404

• A new function named seq_space() has been added. This function is similiar to diff(), supporting calculated the lagged and iterated differences.
• A new function named seq_reverse() has been added. This function can return the vector in reversed relative order.
• A new function named seq_evermax() has been added. This function can return whether the number is the maximum number in the sub-vector from the beginning to the current position.

20200117

• A new function named MixedModelWrite() has been added. This function support writing the fixed effects of a given model and the descriptive information of a given data into your disk. This will help you tidy the results in your paper. In addition, it will generate the qqplot of the residuals of the model.
• A new function named MixedModelExplore() has been added. This function can offer the information about the distribution of the dependent variable. This will help you detect whether it is needed to delete some value (e.g. outliers) or to transform the dependent vaiable (e.g. log-transformed).

20200110

• MixedModelOpt() Can offer suggestions if the there is any redundant random factor. Users can choose whether the contrasts manner of the fixed factors has been set via parameter ContrastsM; set it TRUE means the contrasts manner has been set mannually; otherwise, the function will use contr.simple() to set. Remember that the output will only report the existence of the redundant random factor rather than point out it/them for now. So which random factor(s) is/are redundant should be checked further.

• MixedModelPlot() For violin-plus-raw-data plot, users can choose whether to use the mean of DV across any group variable such as subject or item rather than use the raw observations via set the parameter Group; if the parameter is set NULL (this is also the default value), raw observations will be used to plot; otherwise, if users set it as the name of a given group variable, the mean values of raw observations on each unit in this group will be first calculated, and then the mean values are used to plot.