Natural feeling domain-specific language for building structural equation models in R for estimation by covariance-based methods (like LISREL/Lavaan) or partial least squares (like SmartPLS)
@NicholasDanks: vignettes
@soumyaray: README
both: demos
User has reported one combination of model+data that causes bootstrapping to fail for nboot >= 300. Contact @soumyaray for example of model+data
I was able to reproduce the error the user reported during repeated bootstrapping tests:
> boot_test_pls <- bootstrap_model(seminr_model = cm_pls, nboot = 400)
Bootstrapping model using seminr...
Bootstrapping encountered this ERROR:
2 nodes produced errors; first error: system is computationally singular: reciprocal condition number = 0
On rare occasions, I also encountered:
> boot_test_pls <- bootstrap_model(seminr_model = cm_pls, nboot = 400)
Bootstrapping model using seminr...
Bootstrapping encountered this ERROR:
one node produced an error: missing value where TRUE/FALSE needed
Error does not seem to occur for nboot=100, but rather for nboot=300 or 400
Setting bootstrapping seed (e.g., set.seed(42)) seems to resolve the issue, which leads me to suspect that there are problematic resamples involved.
We have to find a particular resample that causes the problem and identify what is going wrong. If it is truly the fault of the resampled data, we might consider disposing of bad resamples.
Following warning shows up even if model does not include reflective constructs:
Models with interactions cannot be estimated as PLS consistent
and therefore no adjustment for PLS consistent has been made
For a PLS consistent simple interaction model please refer to PLSc_interact() function
Example: run demo/seminr_contained.R
Should only show up for models with reflective constructs involved in interactions
Able to generate hypotheses testing for multi-group analysis (MGA) including measurement invariance as prerequisite for doing MGA
The p and t values for bootstrapped HTMT fluctuate largely (20%) between n = 500 bootstraps
Create a function to automatically generate higher order constructs and integrate them in the model.
Either add construct descriptives in summary, or show how to get them in the vignette/manuals
Add Juan as aut in DESCRIPTION file
ltVariables —> construct_names
mimic folders in R/ with filenames:
estimate_*.Rassess_*.Rcompute_metrics.R (specific stats formula will be performed here)duplicate rho_a (evaluation.R) and rhoA (reliability.R)
remove "SEMinR" title from README.md — the logo suffices as the title
add TravisCI badge (build passing) to README.md
should TravisCI badge link to Travis build page?
change interaction item names from con1.con2 to con1*con2
compute_vif() function cannot handle construct names with a space. eg "Intention Purchase" is not accepted, but "IntentionPurchase" is accepted.
This bug arises from the use of paste() in the formula for lm
compute_vif <- function(target, predictors, model_data) {
independents_regr <- stats::lm(paste(target," ~."),
data = as.data.frame(model_data[,predictors]))
r_squared <- summary(independents_regr)$r.squared
1/(1 - r_squared)
}
Add a test for constructs with spaces in name
A windows firewall pop-up warning occurred during running parallel function in bootstrap_model().
https://stackoverflow.com/questions/47353848/localhost-connection-without-firewall-popup/47491842
Should we apply factor or composite evaluation to interaction terms?
Perhaps copy over dataset or use new datasaet
#### FIRST MODEL 1 ####
# Creating our measurement model
model1_mm <- constructs(
reflective("IP", multi_items("X", 1:21)),
composite("CO", single_item("PECI"), weights = mode_A),
reflective("Acquiescence", multi_items("AQ", 1:3)),
reflective("Compromise", multi_items("COMP", 1:3)),
reflective("Avoid", multi_items("AVO", 1:3)),
reflective("Defy", multi_items("DEF", 1:4)),
reflective("Manipulate", multi_items("MAN", 1:3)),
reflective("Control", single_item("Control_Var_2"))
)
# Interaction constructs must be created after the measurement model is defined.
# We are using the orthogonalization method as per Henseler & Chin (2010)
model1_xm <- interactions(
interaction_ortho("IP", "CO"))
# Structural model
# note: interactions should be the names of its main constructs joined by a '*' in between.
model1_sm <- relationships(
paths(from = "IP", to = c("Acquiescence", "Compromise", "Avoid", "Defy", "Manipulate")),
paths(from = "CO", to = c("Acquiescence", "Compromise", "Avoid", "Defy", "Manipulate")),
paths(from = "Control", to = c("Acquiescence", "Compromise", "Avoid", "Defy", "Manipulate")),
paths(from = "IP*CO", to = c("Acquiescence", "Compromise", "Avoid", "Defy", "Manipulate")))
# Load data, assemble model, and estimate using simplePLS
model1_pls <- estimate_pls(data = data,
measurement_model = model1_mm,
interactions = model1_xm,
structural_model = model1_sm,
inner_weights = path_weighting)
Gives this error
Generating the seminr model
All 96 observations are valid.
Error in mm[mm[, "construct"] == construct2, ][, "measurement"] :
incorrect number of dimensions
this is my names(data)
names(data)
[1] "Response_ID" "IP_Address" "Timestamp" "Duplicate" "Time_Taken"
[6] "Seq_Number" "Country_Code" "Region" "Response_Status" "Browser"
[11] "Device" "Operating_System" "Language" "Q45" "Qualificadora"
[16] "X1" "X2" "X3" "X4" "X5"
[21] "X6" "X7" "X8" "X9" "X10"
[26] "X11" "X12" "X13" "X14" "X15"
[31] "X16" "X17" "X18" "X19" "X20"
[36] "X21" "AQ1" "AQ2" "AQ3" "COMP1"
[41] "COMP2" "COMP3" "AVO1" "AVO2" "AVO3"
[46] "DEF1" "DEF2" "DEF3" "DEF4" "MAN1"
[51] "MAN2" "MAN3" "X38" "X39" "X40"
[56] "X41" "X42" "X43" "X44" "X45"
[61] "X46" "X47" "X48" "X49" "X50"
[66] "X51" "X52" "X53" "X54" "X55"
[71] "X56" "X57" "X58" "X59" "X60"
[76] "X61" "X62" "X63" "X64" "X65"
[81] "X66" "X67" "X68" "X69" "X70"
[86] "X71" "X72" "X73" "X74" "X75"
[91] "X76" "X77" "X78" "X79" "X80"
[96] "X81" "X82" "Priority_Time" "Priority_Cost" "Priority_Scope"
[101] "Control_Var_1" "Control_Var_2" "Control_Var_3" "Opinion_Survey" "PECI"
A file of published references that we can refer to specific publications for warnings, etc.
Filename: REFERENCES.yaml / .md
Perhaps some Yaml/md format as:
---
- code: [short code here]
ref: [full academic reference here]
cite: [preferred citation format]
- code: [short code here]
ref: [full academic reference here]
cite: [preferred citation format]
Let's play with some sample bootstrapping notation?
Include a feature that simulates data for a given structural and measurement model. This should simulate both composite and common-factor data.
Use demo scripts for scenarios
Present output tables for CIs for outer weights, outer loadings, indirect paths @daniel McKay "SEMinR - PLS for R" fb group
Measurement model should be specified as:
C = Reflective (mode A consistent)
A = Formative Composite (mode A)
B = Formative Composite (mode B)
It seems our structural model must use every item of our measurement model — must this be the case? It would be nice to create many different structural models from a single measurement model.
E.g.:
library(seminr)
mobi_mm <- constructs(
composite("Image", multi_items("IMAG", 1:5)),
composite("Expectation", multi_items("CUEX", 1:3)),
composite("Quality", multi_items("PERQ", 1:7)),
composite("Value", multi_items("PERV", 1:2)),
composite("Satisfaction", multi_items("CUSA", 1:3)),
composite("Complaints", single_item("CUSCO")),
composite("Loyalty", multi_items("CUSL", 1:3))
)
antecedents <- c("Image", "Expectation", "Quality")
mediators <- c("Satisfaction", "Value")
outcomes <- c("Loyalty", "Complaints")
non_mediated_sm <- relationships(
paths(from = antecedents, to = outcomes)
)
mobi_pls <- estimate_pls(data = mobi,
measurement_model = mobi_mm,
structural_model = non_mediated_sm)
estimate_pls() reports:
Error in warning_struc_meas_model_complete(structural_model, measurement_model, :
The manifest variables must occur as columns in the data.
in estimate_pls() return object change name of interactions from "mobi_xm" to "interactions"
Output:
Models with interactions cannot be estimated as PLS consistent and therefore no adjustment for PLS consistent has been made
I’m getting this message even if no reflective variables were involved in the interaction (main effects or outcome). Shouldn’t I be allowed to have a reflective construct that doesn’t touch the interaction?
dgr[i,1] <- N/A
should be NA?
An issue to help install 'seminr' package.
install.packages('seminr', dependencies=TRUE, type="source")
perhaps SEMinR's DESCRIPTION file needs the Depends: … line to include parallel or any other packages needed. instead of Imports: ...
Leave scores in seminr_model object, but remove them from seminr_model_summary
seminr_model_summary, call the scores object: composite_scoresfunction rhoC_AVE has problem here:
if(measure_mode(i,seminr_model$mmMatrix)=="B"| measure_mode(i,seminr_model$mmMatrix)=="A"){
x <- seminr_model$outer_loadings[, i]
ind <- which(x!=0)
if(length(ind)==1){
dgr[i,1:2] <- 1
} else {
x <- x[ind]
dgr[i,1] <- sum(x)^2 / (sum(x)^2 + sum(1-x^2))
dgr[i,2] <- sum(x^2)/length(x)
}
} else {
dgr[i,1] <- N/A
dgr[i,2] <- sum(x^2)/length(x)
}
x is defined in if clause but also used in the else clause. this causes it to crash with error if the else clause is reached.
Change the syntax for interacting with SEMinR such that:
Eg. of code:
mobi_mm <- constructs(
composite("Image", multi_items("IMAG", 1:5)),
composite("Expectation", multi_items("CUEX", 1:3)),
composite("Value", multi_items("PERV", 1:2)),
composite("Satisfaction", multi_items("CUSA", 1:3)),
interaction("Image*Expectation", dimensions = c("Image","Satisfaction"), method = two_stage, weights = mode_A),
interaction("Image*Expectation", dimensions = c("Image","Satisfaction"), method = ortho, weights = mode_A),
interaction("Image*Expectation", dimensions = c("Image","Satisfaction"), method = scaled, weights = mode_A),
higher_composite("Value", dimensions = c("Image","Satisfaction"), method = two_stage, weights = mode_A)
)
Composites:
Factors
} else {
dgr[i,1] <- NA
dgr[i,2] <- sum(x^2)/length(x)
should be:
ind <- which(x!=0)
x <- x[ind]
dgr[i,1] <- NA
dgr[i,2] <- sum(x^2)/length(x)
Put on README :)
Either change the bootstrap tests to allow certain variance
Or allow bootstrap to take a seed
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.