PASW Data Preparation
All researchers have to prepare their data prior to analysis. While data preparation tools are included in the PASW Statistics Base product,
sometimes you need more specialized techniques to get your data ready. With the PASW Data Preparation (formerly SPSS Data Preparation) module,
you gain new techniques to help you streamline the data preparation stage of the analytical process. This module helps you to:
- Identify suspicious or invalid cases, variables, and data values
- View patterns of missing data
- Summarize variable distributions
- More accurately get your data ready for analysis
Learn how PASW Data Preparation's techniques can help you get ready for analysis faster and reach more accurate conclusions.
Expand Your Data Preparation Techniques
Use the specialized data preparation techniques in PASW Data Preparation to facilitate data preparation in the analytical process. PASW Data Preparation easily plugs into PASW Statistics Base so you can seamlessly work in the
PASW environment.
Perform Data Checks
Data validation has typically been a manual process. You might run a frequency on your data, print the frequencies, circle what needs to be fixed,
and check for case IDs. Needless to say, this is time consuming. And since every analyst in your organization could use a slightly different method, maintaining consistency from project to project may be a challenge.
To eliminate manual checks, use the Validate Data procedure. This procedure enables you to apply rules to perform data checks based on each variable's
measure level (whether categorical or continuous). For example, if you're analyzing survey data that has variables on a five-point Likert scale, use the
Validate Data procedure to apply a rule for five-point scales and flag all cases that have values outside of the 1-5 range. You can receive reports of invalid
cases as well as summaries of rule violations and the number of cases affected. You can specify validation rules for individual variables (such as range
checks) and cross-variable checks (for example, "pregnant males").
With this knowledge you can determine data validity and remove or correct suspicious cases at your discretion prior to analysis.
Quickly Find Multivariate Outliers
Prevent outliers from skewing analyses when you use the Anomaly Detection procedure. This procedure searches for unusual cases based upon deviations
from similar cases and gives reasons for such deviations. You can flag outliers by creating a new variable. Once you have identified unusual cases, you can
further examine them and determine if they should be included in your analyses.
Preprocess Data Prior to Model Building
In order to use algorithms that are designed for nominal attributes (such as Naïve Bayes and logit models), you must bin your scale variables prior to
model building. If scale variables aren't binned, algorithms such as multinomial logistic regression will take an extremely long time to process or they might
not converge. This is especially true if you have a large dataset. In addition, the results you receive may be difficult to read or interpret.
Optimal Binning, however, enables you to determine cutpoints to help you reach the best possible outcome for algorithms designed for nominal attributes.
With this procedure, you can select from three types of binning for preprocessing data prior to model building:
- Unsupervised: Create bins with equal counts
- Supervised: Take the target variable into account to determine cutpoints. This method is more accurate than unsupervised; however, it is also more computationally intensive.
- Hybrid approach: Combines the unsupervised and supervised approaches. This method is particularly useful if you have a large amount of distinct values.
Back to PASW Statistics Base page.
|
