analyzeSGP(sgp_object,
         state=NULL,
         years=NULL,
         content_areas=NULL,
         grades=NULL,
         sgp.percentiles=TRUE,
         sgp.projections=TRUE,
         sgp.projections.lagged=TRUE,
         sgp.percentiles.baseline=TRUE,
         sgp.projections.baseline=TRUE,
         sgp.projections.lagged.baseline=TRUE,
         sgp.percentiles.baseline.max.order=3,
         sgp.percentiles.srs.baseline.max.order=3,
         sgp.projections.baseline.max.order=3,
         sgp.projections.lagged.baseline.max.order=3,
         sgp.projections.max.forward.progression.years=3,
         sgp.projections.max.forward.progression.grade=NULL,
         sgp.projections.use.only.complete.matrices=NULL,
         sgp.minimum.default.panel.years=NULL,
         sgp.use.my.coefficient.matrices=NULL,
         sgp.use.my.sgp_object.baseline.coefficient.matrices=NULL,
         sgp.test.cohort.size=NULL,
         return.sgp.test.results=FALSE,
         simulate.sgps=TRUE,
         calculate.simex=NULL,
         calculate.simex.baseline=NULL,
         calculate.simex.srs.baseline=NULL,
         calculate.srs=NULL,
         calculate.srs.baseline=NULL,
         goodness.of.fit.print=TRUE,
         sgp.config=NULL,
         sgp.config.drop.nonsequential.grade.progression.variables=TRUE,
         sgp.baseline.panel.years=NULL,
         sgp.baseline.config=NULL,
         trim.sgp.config=TRUE,
         parallel.config=NULL,
         verbose.output=FALSE,
         print.other.gp=NULL,
         sgp.projections.projection.unit="YEAR",
         get.cohort.data.info=FALSE,
         sgp.sqlite=FALSE,
         sgp.percentiles.equated=NULL,
         sgp.percentiles.equating.method=NULL,
         sgp.percentiles.calculate.sgps=TRUE,
         SGPt=NULL,
         fix.duplicates=NULL,
         ...)

Arguments

sgp_object

An object of class SGP containing long formatted data in the @Data slot (from prepareSGP).

state

Acronym indicating state associated with the data for access to embedded knot and boundaries, cutscores, CSEMs, and other state related assessment data.

years

A vector indicating year(s) in which to produce student growth percentiles and/or student growth projections/trajectories. If missing the function will use the data to infer the year(s) based upon the assumption of having at least three years of panel data for analyses.

content_areas

A vector indicating content area(s) in which to produce student growth percentiles and/or student growth projections/trajectories. If left missing the function will use the data to infer the content area(s) available for analyses.

grades

A vector indicating grades for which to calculate student growth percentiles and/or student growth projections/trajectories. If left missing the function will use the data to infer all the grade progressions for student growth percentile and student growth projections/trajectories analyses.

sgp.percentiles

Boolean variable indicating whether to calculate student growth percentiles. Defaults to TRUE.

sgp.projections

Boolean variable indicating whether to calculate student growth projections. Defaults to TRUE.

sgp.projections.lagged

Boolean variable indicating whether to calculate lagged student growth projections often used for growth to standard analyses. Defaults to TRUE.

sgp.percentiles.baseline

Boolean variable indicating whether to calculate baseline student growth percentiles and/or coefficient matrices. Defaults to TRUE.

sgp.projections.baseline

Boolean variable indicating whether to calculate baseline student growth projections. Defaults to TRUE.

sgp.projections.lagged.baseline

Boolean variable indicating whether to calculate lagged baseline student growth projections. Defaults to TRUE.

sgp.percentiles.baseline.max.order

Integer indicating the maximum order to calculate baseline student growth percentiles (regardless of maximum coefficient matrix order). Also the max order of baseline coefficient matrices to be calculated if requested. Default is 3. To utilize the maximum matrix order, set to NULL.

sgp.percentiles.srs.baseline.max.order

Integer indicating the maximum order to calculate baseline, stratified random sample, student growth percentiles (regardless of maximum coefficient matrix order). Also the max order of baseline SRS coefficient matrices to be calculated if requested. Default is 3. To utilize the maximum matrix order, set to NULL.

sgp.projections.baseline.max.order

Integer indicating the maximum order to calculate baseline student growth projections (regardless of maximum coefficient matrix order). Default is 3. To utilize the maximum matrix order, set to NULL.

sgp.projections.lagged.baseline.max.order

Integer indicating the maximum order to calculate lagged baseline student growth projections (regardless of maximum coefficient matrix order). Default is 3. To utilize the maximum matrix order, set to NULL.

sgp.projections.max.forward.progression.years

Integer indicating the maximum number of years forward that cohort based projections will be established for. Default is 3 years.

sgp.projections.max.forward.progression.grade

Integer indicating the maximum grade forward that cohort based projections will be established for. Default is NULL, the highest grade.

sgp.projections.use.only.complete.matrices

Boolean argument (defaults to TRUE/NULL) indicating whether to produce projections only when a complete set of coefficient matrices is available.

sgp.minimum.default.panel.years

Integer indicating the minimum number of panels years to use for default sgp analyses. Default value is NULL (converted to 3) years of data.

sgp.use.my.coefficient.matrices

Argument, defaults to NULL, indicating whether to use coefficient matrices embedded in argument supplied to 'sgp_object' to calculate student growth percentiles.

sgp.use.my.sgp_object.baseline.coefficient.matrices

Argument, defaults to NULL (FALSE), indicating whether to utilize baseline matrices embedded in supplied sgp_object and not utilize baseline matrices embedded in SGPstateData.

sgp.test.cohort.size

Integer indicating the maximum number of students sampled from the full cohort to use in the calculation of student growth percentiles. Intended to be used as a test of the desired analyses to be run. The default, NULL, uses no restrictions (no tests are performed, and analyses use the entire cohort of students).

return.sgp.test.results

Boolean variable passed to studentGrowthPercentiles indicating whether the results from the cohort sample subset (if specified using the above argument) should be returned for inspection. Defaults to FALSE. If TRUE, only the sample subset of the data used will be returned in the SGP object's @Data slot. Alternatively, user can supply the character "ALL_DATA" to the argument to return the entire original data.

simulate.sgps

Boolean variable indicating whether to simulate SGP values for students based on test-specific Conditional Standard Errors of Measurement (CSEM). Test CSEM data must be available for simulation and included in SGPstateData. This argument must be set to TRUE for confidence interval construction. Defaults to TRUE.

calculate.simex

A character state acronym or list including state/csem variable, csem.data.vnames, csem.loss.hoss, simulation.iterations, lambda and extrapolation method. Returns both SIMEX adjusted SGP (SGP_SIMEX) as well as the percentile ranked SIMEX SGP (RANK_SIMEX) values as suggested by Castellano and McCaffrey (2017). Defaults to NULL, no simex calculations performed. Alternatively, setting the argument to TRUE sets the list up with state=state, lambda=seq(0,2,0.5), simulation.iterations=50, simex.sample.size=25000, extrapolation="linear" and save.matrices=TRUE.

calculate.simex.baseline

A character state acronym or list including state/csem variable, csem.data.vnames, csem.loss.hoss, simulation.iterations, lambda and extrapolation method. Defaults to NULL, no simex calculations performed. Alternatively, setting the argument to TRUE uses the same defaults as above along with simex.use.my.coefficient.matrices = TRUE, which assumes baseline SIMEX coefficient matrices are available.

calculate.simex.srs.baseline

A character state acronym or list including state/csem variable, csem.data.vnames, csem.loss.hoss, simulation.iterations, lambda and extrapolation method. Defaults to NULL, no simex calculations performed for stratified random sample (SRS). Alternatively, setting the argument to TRUE uses the same defaults as above along with simex.use.my.coefficient.matrices = TRUE, which assumes baseline SIMEX coefficient matrices are available.

calculate.srs

A character state acronym or list including [FILL IN LATER]. Creates a longitudinal data set based upon a stratified random sample of variables and proportions for the United States (default) or provided by the user. The argument defaults to NULL, Alternatively, setting the argument to TRUE uses the defaults specified above.

calculate.srs.baseline

A character state acronym or list including [FILL IN LATER]. Calculates SGPs based upon previosly established coefficient matrices derived from a stratified random sample of data. Defaults to NULL, no stratified random sample SGPs are calculated. Alternatively, setting the argument to TRUE uses the defaults specified above.

goodness.of.fit.print

Boolean variable indicating whether to print out Goodness of Fit figures as PDF into a directory labeled Goodness of Fit. Defaults to TRUE.

sgp.config

If years, content_areas, and grades are missing, user can directly specify a list containing three vectors: baseline.content.areas, baseline.panel.years, and baseline.grade.sequences. This advanced option is helpful for analysis of non-traditional grade progressions and other special cases. See examples for use cases.

sgp.config.drop.nonsequential.grade.progression.variables

Boolean variable (defaults to TRUE) indicating whether non-sequential grade progression variables should be dropped when sgp.config is processed. For example, if a grade progression of c(3,4,6) is provided, the data configuration will assume (default is TRUE) that data for a missing year needs to be dropped prior to applying studentGrowthPercentiles or studentGrowthProjections to the data.

sgp.baseline.panel.years

A vector of years to be used for baseline coefficient matrix calculation. Default is to use most recent five years of data.

sgp.baseline.config

A list containing three vectors: sgp.content.areas, sgp.panel.years, sgp.grade.sequences indicating how baseline student growth percentile analyses are to be conducted. In almost all cases this value is calculated by default within the function but can be specified directly for advanced use cases. See source code for more detail on this configuration option.

trim.sgp.config

A Boolean variable indicating whether the arguments content_areas, years and grades should be used to 'trim' any manually supplied configuration for analysis supplied by 'sgp.config'.

parallel.config

A named list with, at a minimum, two elements indicating 1) the BACKEND package to be used for parallel computation and 2) the WORKERS list to specify the number of processors to be used in each major analysis. The BACKEND element can be set = to FOREACH or PARALLEL. Please consult the manuals and vignettes for information of these packages!

TYPE is a third element of the parallel.config list that provides necessary information when using FOREACH or PARALLEL packages as the backend. With BACKEND="FOREACH", the TYPE element specifies the flavor of 'foreach' backend. As of version 1.0-1.0, only "doParallel" is supported. If BACKEND = "PARALLEL", the parallel package will be used. This package combines deprecated parallel packages snow and multicore. Using the "snow" implementation of parallel the function will create a cluster object based on the TYPE element specified and the number of workers requested (see WORKERS list description below). The TYPE element indicates the users preferred cluster type (either "PSOCK" for socket cluster of "MPI" for an OpenMPI cluster). If Windows is the operating system, this "snow" implementation must be used and the TYPE element must = "PSOCK". Defaults are assigned based on operating system if TYPE is missing based on system OS. Unix/Mac OS defaults to the "multicore" to avoid worker node pre-scheduling and appears to be more efficient in these operating systems.

The WORKERS list must contain, at a minimum, a single number of processors (nodes) desired or available. If WORKERS is specified in this manner, then the same number of processors will be used for each analysis type (sgp.percentiles, sgp.projections, ... sgp.projections.lagged.baseline). Alternatively, the user may specify the numbers of processors used for each analysis. This allows for better memory management in systems that do not have enough RAM available per core. The choice of the number of cores is a balance between the number of processors available, the amount of RAM a system has and the size of the data (sgp_object). Each system will be different and will require some tailoring. One rule of thumb used by the authors is to allow for 4GB of memory per core used for running large state data. The SGP Demonstration (and data that size) requires more like 1-2GB per core. As an example, PERCENTILES=4 and PROJECTIONS=2 might be used on a quad core machine with 4 GB of RAM. This will use all 4 cores available for the sgp.percentiles analysis and 2 cores for the sgp.projections analysis (which requires more memory than available). The WORKERS list accepts these elements: PERCENTILES, PROJECTIONS (for both cohort and baseline referenced projections), LAGGED_PROJECTIONS (for both cohort and baseline referenced lagged projections), BASELINE_MATRICES (used to produce the baseline coefficient matrices when not available in SGPstateData - very computationally intensive), BASELINE_PERCENTILES (SGP calculation only when baseline coefficient matrices have already been produced and are available - NOT very computationally intensive).

Alternatively, the name of an external CLUSTER.OBJECT (PSOCK or MPI) set up by the user outside of the function can be used.

Example use cases are provided below.

verbose.output

A Boolean argument (defaults to FALSE) indicating whether the function should output verbose diagnostic messages.

print.other.gp

A Boolean argument (defaults to FALSE) indicating whether the function should output SGP of all orders.

sgp.projections.projection.unit

A character vector argument indicating whether the studentGrowthProjections function should produce projections relative to future grades or future years. Options are "YEAR" and "GRADE", with default being "YEAR".

get.cohort.data.info

A Boolean argument (defaults to FALSE) indicating whether a summary of all cohorts to be submitted to the studentGrowthPercentiles and studentGrowthProjections functions should be performed prior to analysis.

sgp.sqlite

A Boolean argument (defaults to FALSE) indicating whether a SQLite database file of the essential SGP data should be created from the @Data slot and subsequently used to extract data subsets for analysis with studentGrowthPercentiles and studentGrowthProjections functions. If the size of the @Data object is greater than 1 GB sgp.sqlite is set to TRUE internally. When TRUE, this can substantially reduce the amount of RAM memory required to conduct analyses. If set to TRUE the file "TMP_SGP_Data.sqlite" will be created in the R temporary directory (see?tempdir for information). This file is deleted by default although one may keep it if the argument is specified as the character "KEEP".

sgp.percentiles.equated

A Boolean argument (defaults to NULL/FALSE) indicating whether equating should be used on the most recent year of test data provided. Equating allows for student growth projections to be calculated in across assessment transitions where the scale for the assessment changes.

sgp.percentiles.equating.method

A character vector (defaults to NULL/'equipercentile') indicating the type of equating method to use. Options include any combination of 'identity', 'mean', 'linear', and 'equipercentile'.

sgp.percentiles.calculate.sgps

A Boolean argument (defaults to TRUE) indicating whether to calculate percentiles in calls to studentGrowthPercentiles function. Setting to FALSE would indicate desire to calculate only coefficient matrices and no percentiles.

SGPt

An argument supplied to implement time-dependent SGP analyses (SGPt). Default is NULL giving standard, non-time dependent argument. If set to TRUE, the function assumes the variables 'TIME' and 'TIME_LAG' are supplied as part of the panel.data. To specify other names, supply a list of the form: list(TIME='my_time_name', TIME_LAG='my_time_lag_name'), substituting your variable names.

fix.duplicates

Argument to control how duplicate records based upon the key of VALID_CASE, CONTENT_AREA, YEAR, and ID are dealt with. If set to 'KEEP.ALL', the function tries to fix the duplicate individual records by adding a '_DUP_***' suffix to the duplicate ID before running studentGrowthPercentiles in order to create unique records based upon the key. See combineSGP for additional info on fix.duplicates functionality.

...

Arguments to be passed to studentGrowthPercentiles or studentGrowthProjections for finer control over SGP calculations. NOTE: arguments can only be passed to one lower level function at a time, and only student growth percentiles OR projections can be created but not both at the same time.

Value

Function returns a list containing the long data set in the @Data slot as a data.table keyed using VALID_CASE, CONTENT_AREA, YEAR, ID and the student growth percentile and/or student growth projection/trajectory results in the SGP slot.

See also

prepareSGP, combineSGP

Examples

not_run({ ## analyzeSGP is Step 2 of 5 of abcSGP Demonstration_SGP <- sgpData_LONG Demonstration_SGP <- prepareSGP(Demonstration_SGP) Demonstration_SGP <- analyzeSGP(Demonstration_SGP) ## Or (explicitly pass state argument) Demonstration_SGP <- prepareSGP(sgpData_LONG) Demonstration_SGP <- analyzeSGP(Demonstration_SGP, state="DEMO") ### ### Example uses of the sgp.config argument ### # Use only 3 years of Data, for grades 3 to 6 # and only perform analyses for most recent year (2012) my.custom.config <- list( MATHEMATICS.2013_2014 = list( sgp.content.areas=rep("MATHEMATICS", 3), # Note, must be same length as sgp.panel.years sgp.panel.years=c('2011_2012', '2012_2013', '2013_2014'), sgp.grade.sequences=list(3:4, 3:5, 4:6)), READING.2013_2014 = list( sgp.content.areas=rep("READING", 3), sgp.panel.years=c('2011_2012', '2012_2013', '2013_2014'), sgp.grade.sequences=list(3:4, 3:5, 4:6))) Demonstration_SGP <- prepareSGP(sgpData_LONG) Demonstration_SGP <- analyzeSGP(Demonstration_SGP, sgp.config=my.custom.config, sgp.percentiles.baseline = FALSE, sgp.projections.baseline = FALSE, sgp.projections.lagged.baseline = FALSE, simulate.sgps=FALSE) ## Another example sgp.config list: # Use different CONTENT_AREA priors, and only 1 year of prior data my.custom.config <- list( MATHEMATICS.2013_2014.READ_PRIOR = list( sgp.content.areas=c("READING", "MATHEMATICS"), sgp.panel.years=c('2012_2013', '2013_2014'), sgp.grade.sequences=list(3:4, 4:5, 5:6)), READING.2013_2014.MATH_PRIOR = list( sgp.content.areas=c("MATHEMATICS", "READING"), sgp.panel.years=c('2012_2013', '2013_2014'), sgp.grade.sequences=list(3:4, 4:5, 5:6))) ## An example showing multiple priors within a single year Demonstration_SGP <- prepareSGP(sgpData_LONG) DEMO.config <- list( READING.2012_2013 = list( sgp.content.areas=c("MATHEMATICS", "READING", "MATHEMATICS", "READING", "READING"), sgp.panel.years=c('2010_2011', '2010_2011', '2011_2012', '2011_2012', '2012_2013'), sgp.grade.sequences=list(c(3,3,4,4,5), c(4,4,5,5,6), c(5,5,6,6,7), c(6,6,7,7,8))), MATHEMATICS.2012_2013 = list( sgp.content.areas=c("READING", "MATHEMATICS", "READING", "MATHEMATICS", "MATHEMATICS"), sgp.panel.years=c('2010_2011', '2010_2011', '2011_2012', '2011_2012', '2012_2013'), sgp.grade.sequences=list(c(3,3,4,4,5), c(4,4,5,5,6), c(5,5,6,6,7), c(6,6,7,7,8)))) Demonstration_SGP <- analyzeSGP( Demonstration_SGP, sgp.config=DEMO.config, sgp.projections=FALSE, sgp.projections.lagged=FALSE, sgp.percentiles.baseline=FALSE, sgp.projections.baseline=FALSE, sgp.projections.lagged.baseline=FALSE, sgp.config.drop.nonsequential.grade.progression.variables=FALSE) ### ### Example uses of the parallel.config argument ### ## Windows users must use a snow socket cluster: # possibly a quad core machine with low RAM Memory # 4 workers for percentiles, 2 workers for projections. # Note the PSOCK type cluster is used for single machines. Demonstration_SGP <- prepareSGP(sgpData_LONG) Demonstration_SGP <- analyzeSGP(Demonstration_SGP, parallel.config=list( BACKEND="PARALLEL", TYPE="PSOCK", WORKERS=list(PERCENTILES=4, PROJECTIONS=2, LAGGED_PROJECTIONS=2, BASELINE_PERCENTILES=4)) ## New parallel package - only available with R 2.13 or newer # Note there are up to 16 workers, and MPI is used, # suggesting this example is for a HPC cluster, possibly Windows OS. ... parallel.config=list( BACKEND="PARALLEL", TYPE="MPI", WORKERS=list(PERCENTILES=16, PROJECTIONS=8, LAGGED_PROJECTIONS=6, BASELINE_PERCENTILES=12)) ... ## FOREACH use cases: ... parallel.config=list( BACKEND="FOREACH", TYPE="doParallel", WORKERS=3) ... # NOTE: This list of parallel.config specifications is NOT exhaustive. # See examples in analyzeSGP documentation for some others.0 ### ### Advanced Example: restrict years, recalculate baseline SGP ### coefficient matrices, and use parallel processing ### # Remove existing DEMO baseline coefficient matrices from # the SGPstateData object so that new ones will be computed. SGPstateData$DEMO$Baseline_splineMatrix <- NULL # set up a customized sgp.config list . . . # set up a customized sgp.baseline.config list . . . # to be completed })
#> Error: <text>:98:9: unexpected symbol #> 97: # suggesting this example is for a HPC cluster, possibly Windows OS. #> 98: ... #> ^