Missing data analysis was conducted and it was determined that there was no
significant pattern to the missing data and data was missing at random. A multiple
imputation technique was used to address missing data. Five imputations of the missing
data were created so in order to introduce the appropriate random error and create a more
accurate estimate of values. Values were imputed based on the predictive variables of:
the age the student was first identified as having a disability, and what type of services
they qualify for, specifically if they qualify for a 504 Plan or tutoring. Separate multiple
imputations of missing data were run on the subset of the students who attended college.
In this sample, household income, and the type of post-secondary institution they
attended were used to predict missing values. This subset was used in the analysis of
persistence. No dependent variables were used as predictors in either multiple
77
imputation set so as not to bias the results. Missing values were not computed for
dependent variables. Regressions were run using the aggregate value for the original data
and the five imputed values to provide the most accurate approximation of a complete
dataset. Due to limitations in the statistical software used (SPSSS Missing Values) the
pooled results of the original data and the five imputations could not be used to compute
the percentage of variance explained in the logistic regressions or the R-square in the
linear regression. The values for the first imputation were used because they were the
most moderate outcomes of the different imputations, though no significant difference
was present in any of the imputations. The pooled data was used for the regression
coefficients and all other analyses.
The second phase of the data analysis involves the use of descriptive statistics (e.g.
univariate frequency distributions, crosstabs, and comparison of means) to examine
patterns of enrollment and persistence in college students with learning disabilities, and
in the general population. The analysis will be run on the NTLS-2 and ELS databases.
The third phase of analysis will examine what factors best predict college enrollment and
persistence. The specific binary outcomes that will be assessed are: attending any form
of post-secondary education or not, attending a two-year college or not, and attending a
four-year college or not. Logistic regressions will be utilized to examine the relationship
between forms of capital and the likelihood of attending any post-secondary institution,
and two-year and four-year institutions, controlling for gender, ethnicity, and academic
performance.