Recap of basic SPSS and statistics

5th - 9th December 2011, Rome

Manage the database Import / export file Import variable from another database / merge files Restructure cases to variables

Merging datasets For each level of investigation in a survey, there is

typically a dataset For example if a survey asks questions at the household

level, then measures anthropometry of children under 5 and for women of reproductive age as well as a community level questionnaire, we would expect 4 separate datasets to be created

To do analysis that looks at a case in respect to its context, datasets must be merged

Merging datasets For example – the education level of a household head is recorded in the

household dataset. We may be interested to find if the nutritional status of a child is related to education of the household head. But the child data is in a separate dataset.

In order to merge the datasets, a common variable must exist in each dataset. In this case, a household identifier must be in both datasets.

Household ID

Education level of household head

Household ID

Weight for Age z-score (WAZ)

Household dataset Child dataset

Merging datasets In each dataset, the cases must be sorted on the household identifier In SPSS, select Merge Files > Add Variables; select the datasets and the

variable to match the datasets on The new dataset will have the variable of interest included; In our example,

we will now have a child dataset that also has the literacy of the education level of the household head included and can do our analysis

Household ID

Education level of household head

Household ID

Weight for Age z-score (WAZ)

Household dataset Child datasetHousehold ID

Weight for Age z-score (WAZ)

Education level of household head

Child dataset +

Data cleaning

Unique ID Missing data Define variable properties Scatteplot /histograms Frequency sorting Outliers

Missing values and data cleaning Cleaning data can be a painful process Being systematic about cleaning data from the beginning

of the process can save hours of work later in the analysis There are few key tools to use in SPSS to clean data:

Sorting cases – allows you to quickly see within a variable if there are problematic cases

Indentify duplicate cases – shows cases which have the same unique identifier

Histograms and scatterplots – visually identify problematic variables and cases

Missing values and data cleaning The data cleaning process will also reveal cases where

values are missing for certain variables This is often expected (though in some cases may have

been an error) Handling missing values in SPSS is a simple matter of

telling the software what values to handle as missing in the variable view

Getting ready for analysis Weight file Split file Select cases

Analysis Create new variables

Recode Count Compute Rank cases (quintiles) Aggregate

Frequencies Compare mean Crosstabs

Create new variables using recode Recoding a variable is most commonly

used to take a categorical variable and to re-categorize it’s values.

For example – source of drinking water is a standard question in household surveys with several options that are adapted for the local context.

When describing water sources in analysis, we usually will compare improved vs. unimproved water sources

In the example on the right, the top box represents a module in the household questionnaire and the bottom box represents the categorization of improved vs. unimproved water sources. If we want to recode the question responses into a bi-variate variable, how do we do so in SPSS?

4.2 What is the main source of drinking water for your household? (Circle one)

•1 = Piped water•2 = Well (protected)•3 = Well (unprotected)•4 = River, stream or pond•5 = Collecting rainwater•6 = Tanker truck water

Improved source Unimproved source

Piped water Well (unprotected)

Well (protected) River, stream or pond

Collecting rainwater Tanker truck water

Creating a new variable using compute Computing a new variable is usually

done when a mathematical formula is used to derive a new variable

A number of circumstances in a household questionnaire require computation

For example – a commonly used indicator in assessments when discussing demographics is the percentage of dependents in a household

Given the household questionnaire roster on the right, how can we create a variable for the percentage of dependents (where dependents are people under 15 and over 65)?

1.5 Please complete the household demographics table on the right. Record the number of individuals in each age category, differentiated by males and females.

Age Male Female

a. 0-5 years

|__|__| |__|__|

b. 6-14 years

|__|__| |__|__|

c. 15-64 years

|__|__| |__|__|

d. 65 years or older

|__|__| |__|__|

Type of variables

Continuous(Scale)

Categorical

Interval ex. Age 1 to n

Ratioex. Percentage of expenditure 0% to 100%

Nominal The categories are not ranked ex. 1=female, 2=male

Ordinal The categories are ranked ex. 1=poor, 2=medium, 3= good

We work with two types of variables

Type of variables Type of variables

Quantitative

Qualitative

Ordinal Nominal

Do arithmetic operations on values make sense?

Yes No No

Are values ordered? Yes Yes No

Types of values Numeric Alphanumeric codes

Alphanumeric codes

Descriptive statistics

Continuous Categorical

RangeMean Median Mode

FrequenciesCrosstabs

Best practices Using syntax Export files/outputs Data file comments