--- title: "SGP Data Preparation" author: "Damian W Betebenner & Adam R Van Iwaarden" date: "September 26th, 2017" output: rmarkdown::html_vignette vignette: > %\VignetteIndexEntry{Data Preparation} %\VignetteEngine{knitr::rmarkdown} %\usepackage[utf8]{inputenc} --- ```{r include = FALSE} library(SGP) library(SGPdata) is_html_output = function() { knitr::opts_knit$get("rmarkdown.pandoc.to")=="html" } knitr::opts_chunk$set( collapse=TRUE, comment="", prompt=TRUE, fig.dpi=96) if (is_html_output()) { options(width=1000) } ``` # Introduction There a two common formats for representing longitudinal (time dependent) student assessment data: WIDE and LONG format. For WIDE format data, each case/row represents a unique student and columns represent variables associated with the student at different times. For LONG format data, time dependent data for the student is spread out across multiple rows in the data set. The [SGPdata package](https://centerforassessment.github.io/SGPdata/), installed when one installs the SGP package, includes exemplar WIDE and LONG data sets (sgpData and sgpData_LONG, respectively) to assist in setting up your data. Deciding whether to format in WIDE or LONG format is driven by many conditions. In terms of the analyses that can be performed using the SGP package, the WIDE data format is used by the lower level functions `studentGrowthPercentiles` and `studentGrowthProjections` whereas the higher level wrapper functions utilize the LONG data format. For all but the simplest, one-off, analyses, you're likely better off formatting your data in the LONG format and using the higher level functions. This is particularly true is you plan on running SGP analyses operationally year after year where LONG data has numerous preparation and storage benefits over WIDE data. # WIDE data format: sgpData Longitudinal data in WIDE format is usually the most "intuitive" longitudinal format for those new to longitudinal/time-dependent data. Each row of the data set provides all the data for the individual case with the variable names indicating what time period the data is from. Though intuitive, the data is often difficult to work with, particularly in situations where data is frequently added to the The data set `sgpData` is an anonymized, panel data set comprisong 5 years of annual, vertically scaled, assessment data in WIDE format. This exemplar data set models the format for data used with the lower level [`studentGrowthPercentiles`](https://sgp.io/reference/studentGrowthPercentiles.html) and [`studentGrowthProjections`](https://sgp.io/reference/studentGrowthProjections.html) functions. ```{r} head(sgpData) ``` The Wide data format illustrated by `sgpData` and utilized by the SGP package can accomodate any number of occurrences but must follow a specific column order. Variable names are irrelevant, position in the data set is what's important: * The first column *must* provide a unique student identifier. * The next set of columns *must* provide the grade level/time associated with the students assessment occurrences. * The next set of columns *must* provide the numeric scores associated with the students assessment occurrences. In `sgpData` above, the first column, *ID*, provides the unique student identifier. The next 5 columns, *GRADE_2013*, *GRADE_2014*, *GRADE_2015*, *GRADE_2016*, and *GRADE_2017*, provide the grade level of the student assessment score in each of the 5 years. The last 5 columns, *SS_2013*, *SS_2014*, *SS_2015*, *SS_2016*, and *SS_2017*, provide the scale scores associated with the student in each of the 5 years. In most cases the student does not have 5 years of test data so the data shows the missing value (NA). Using wide-format data like `sgpData` with the SGP package is, in general, straight forward. ```{r eval=FALSE} sgp_g4 <- studentGrowthPercentiles( panel.data=sgpData, sgp.labels=list(my.year=2015, my.subject="Reading"), percentile.cuts=c(1,35,65,99), grade.progression=c(3,4)) ``` Please consult the [SGP data analysis vignette](SGP_Data_Analysis.html) for more comprehensive documentation on how to use `sgpData` (and WIDE data formats in general) for SGP analyses. # LONG data format: sgpData_LONG The data set `sgpData_LONG` is an anonymized, panel data set comprising 5 years of annual, vertcially scaled, assessment data in LONG format for two content areas (ELA and Mathematics). This exemplar data set models the format for data used with the higher level functions [`abcSGP`](https://sgp.io/reference/abcSGP.html), [`prepareSGP`](https://sgp.io/reference/prepareSGP.html), [`analyzeSGP`](https://sgp.io/reference/analyzeSGP.html), [`combineSGP`](https://sgp.io/reference/combineSGP.html), [`summarizeSGP`](https://sgp.io/reference/summarizeSGP.html), [`visualizeSGP`](https://sgp.io/reference/visualizeSGP.html), and [`outputSGP`](https://sgp.io/reference/outputSGP.html) ```{r} head(sgpData_LONG) ``` We recommend LONG formated data for use with operational analyses. Managing data in long format is more simple than data in the wide format. For example, when updating analyses with another year of data, the data is appended onto the bottom of the currently existing long data set. All higher level functions in the SGP package are designed for use with LONG format data. In addition, these functions often assume the existence of state specific meta-data in the embedded [SGPstateData](https://centerforassessment.github.io/SGPstateData/) meta-data. See the [SGP package documentation](https://sgp.io/reference/index.html) for more comprehensive documentation on how to use `sgpData` for SGP calculations. There are 7 *required* variables when using LONG data with SGP analyses: `VALID_CASE`, `CONTENT_AREA`, `YEAR`, `ID`, `SCALE_SCORE`, `GRADE` and `ACHIEVEMENT_LEVEL` (on required if running student growth projections). `LAST_NAME` and `FIRST_NAME` are required if creating individual level student growth and achievement plots. All other variables are demographic/student categorization variables used for creating student aggregates by the [`summarizeSGP`](https://sgp.io/reference/summarizeSGP.html) function. The `sgpData_LONG` data set contains data for 5 years across 2 content areas (ELA and Mathematics) # LONG data format with time: sgptData_LONG The data set `sgptData_LONG` is an anonymized, panel data set comprising 8 windows (3 windows annually) of assessment data in LONG format for 3 content areas (Early Literacy, Mathematics, and Reading). This data set is similar to the `sgpData_LONG` data set without the demographic variables and with an additional `DATE` variable indicating the date associated with the student assessment record. ```{r} head(sgptData_LONG) ``` # LONG teacher-student lookup: sgpData_INSTRUCTOR_NUMBER The data set `sgpData_INSTRUCTOR_NUMBER` is an anonymized, student-instructor lookup table that provides insturctor information associated with each students test record. Note that just as each teacher can (and will) have more than 1 student associated with them, a student can have more than one teacher associated with their test record. That is, multiple teachers could be assigned to the student in a single content area for a given year. ```{r} head(sgpData_INSTRUCTOR_NUMBER) ``` # Contributions & Requests If you have a contribution or topic request for this vignette, don't hesitate to write or set up an [issue on GitHub](https://github.com/CenterForAssessment/SGP/issues).