19.Multivariate Analysis Using NLTS2 Data
1 Prerequisites Recommended modules to complete before viewing this module 1. Introduction to the NLTS2 Training Modules 2. NLTS2 Study Overview 3. NLTS2 Study Design and Sampling NLTS2 Data Sources, either 4. Parent and Youth Surveys or 5. School Surveys, Student Assessments, and Transcripts 9. Weighting and Weighted Standard Errors
19. Multivariate Analysis Using NLTS2 Data 2 Prerequisites Recommended modules to complete before viewing this module (cont’d) NLTS2 Documentation 10. Overview 11. Data Dictionaries 12. Quick References Accessing Data 14a. Files in SPSS or 14b. Files in SAS 17a. Manipulating Variables in SPSS or 17b. Manipulating Variables in SAS
19. Multivariate Analysis Using NLTS2 Data 3 Overview Multivariate analysis Considerations Variables HLM Critical considerations Closing Important information
19. Multivariate Analysis Using NLTS2 Data 4 Multivariate analysis Explanatory questions Questions regarding relationships among variables, usually requiring controls for covariates Types of multivariate analyses Regression Factor analysis Structural equation model HLM (hierarchical linear modeling) Regression tree
19. Multivariate Analysis Using NLTS2 Data 5 Considerations Select factors that relate to the outcome. Finding the factors that best predict an outcome can be a challenging part of the process. This can become iterative, trying different sets of items for a “best-fit” model. Special considerations for these types of analysis. If any of the items in the model are missing, the respondent may be eliminated from the analysis Variables may need to be recoded for multivariate procedures.
19. Multivariate Analysis Using NLTS2 Data 6 Considerations Missing values If a value is missing, the case may not be included in analysis. Using imputed values is one approach to the problem of missing values. One option is to fill missing values with the mean of the variable by certain characteristics. – For example, reassign the missing value to the mean value of that item using the mean value from those within the same disability category, gender, and age group. Imputed variables should be clearly labeled to differentiate between the original item and the imputed item.
19. Multivariate Analysis Using NLTS2 Data 7 Variables Categorical variables Categorical variables are recoded into a series of variables (also known as “dummy variables”). One variable for each response category. New variables have a value of “1” if that category was indicated and “0” if another category was indicated. One category is omitted from the block. For example, include all disability categories except learning disability (LD) so that it is LD vs. other categories. The program will not run if all categories are included.
19. Multivariate Analysis Using NLTS2 Data 8 Variables Continuous variables Typically no recoding is necessary. Ordinal variables Ordinal variables may be categorical but imply an order. Example: 1 = Never; 2 = Not very often; 3 = Sometimes; 4 = Very often. Ordinal variables can be recoded, as categorical variables. A series of dummies with one category omitted. In the above example, if dummies for values 2, 3, and 4 were included, the ordinal variable would be those who were “never” vs. the other response categories.
19. Multivariate Analysis Using NLTS2 Data 9 Variables Direction of variables The order of responses may have a direction that is either low to high or high to low. A response order such as “(1) never” to “(4) very often” can be either a negative to positive direction or a positive to negative direction based on what question was asked. If the question is “How often do you have trouble getting along with students in your class?” the order would be high to low, with a “(4) very often” being the most negative response and “(1) never” the most positive. If the question is “How often do you get together with friends?” the above order is low to high, with “(4) very often” being the most positive response and “(1) never” the most negative.
19. Multivariate Analysis Using NLTS2 Data 10 Variables Direction of variables (cont’d) If there is a series of questions with the same ordinal response categories, the direction may not be consistent if some of the questions in the series have a negative connotation and others a positive connotation. Reverse-ordering some variables might be worthwhile so that all variables go in the same direction. It can be confusing to interpret positive or negative associations when variables are in mixed directions.
19. Multivariate Analysis Using NLTS2 Data 11 HLM HLM (hierarchical linear model) Multilevel analysis for nested/grouped data For longitudinal studies, time is level A special file needs to be created. HLM requires everything in one file. Data, program, and output HLM variable names are limited to 8 characters. The file should contain only the variables actively used in the model.
19. Multivariate Analysis Using NLTS2 Data 12 Critical considerations Some variable creation and/or file manipulation may be required. Variables that work in other procedures often have to be modified for models. Varying n’s (missing values) have to be considered
19. Multivariate Analysis Using NLTS2 Data 13 Closing Topics discussed in this module Multivariate analysis Considerations Variables HLM Critical considerations Next module: 20. Linear Regression Model: Example
19. Multivariate Analysis Using NLTS2 Data 14 Important information Websites NLTS2 website contains reports, data tables, and other project-related information Information about obtaining the NLTS2 database and documentation can be found on the NCES website General information about restricted data licenses can be found on the NCES website address: