Aggregating Data

Calculating higher-level values

Scratch Variables

Scratch variables are declared with a preceding hash (#). These variables are deleted on reaching execution, and are never re-initialized.

Given sales data, to compute total sales, try:

compute #Total_Sales = sum(Sales, #Total_Sales).
compute Total_Sales = #Total_Sales.

Leave

The LEAVE command explicitly suppresses re-initialization for a given variable. In all other contexts, these are normal variables with normal behavior.

Given sales data, to compute total sales, try:

compute Total_Sales = sum(Sales, Total_Sales).
leave Total_Sales.

Aggregate

The AGGREGATE command can be used to calculate higher-level values.

Given sales data, to compute total sales, try:

aggregate outfile=* mode=addvariables
  /Total_Sales = sum(Sales).

The OUTFILE=* and MODE=ADDVARIABLES subcommands are the default behavior.

Note: if the calculated variables may collide with pre-existing variable names, use the OVERWRITE=YES subcommand.

Functions

See Aggregate Functions.

Break Groups

To calculate a higher-level value for each subgroup, use the /BREAK=VARLIST subcommand. The dataset must be sorted by VARLIST.

If the dataset is already sorted, use the /PRESORTED subcommand to skip re-processing. The /PRESORTED subcommand must precede the /BREAK subcommand.

Aggregating Cases

Aggregate

The AGGREGATE command can be used to create a higher-level dataset.

This is mostly only useful with break groups, whereby a case is generated for each one.

Given state-level sales data, to create region-level sales data, try:

dataset define NEW.
aggregate outfile=NEW
  /break=Region
  /Total_Sales=sum(Sale).

If the dataset is already sorted, use the /PRESORTED subcommand to skip re-processing. The /PRESORTED subcommand must precede the /BREAK subcommand.

Functions

See Aggregate Functions.

Transforming Long to Wide

The CASESTOVARS command transforms data from long to wide format. An index variable identifies the cases that should be aggregated.

Given region-level sales data from multiple years, try:

dataset activate SALES_Y1.
compute Year=1.

dataset activate SALES_Y2.
compute Year=2.

dataset activate SALES_Y3.
compute Year=3.

add files /file=SALES_Y1 /file=SALES_Y2 /file=SALES_Y3.
sort cases by Region.
formats Year (F1).
dataset name SALES.
casestovars
 /id=Region
 /index=Year.

The ID variable (Region) becomes a unique identifier for each row of data. The index variable (Year) is dropped, and all other variables are transformed according to the index variable. For example, the Total_Sales column becomes a set of columns, one for each year (Total_Sales.1, Total_Sales.2, Total_Sales.3).

The index variables can be either numeric or string, but all cases must have a non-missing value.

Constants

If a variable does not vary within the new rows of data, they should instead be treated as constants and should not be transformed. Constants should be specified on the /FIXED subcommand. SPSS will automatically inspect variables and warn about any that appear to be constants but were not specified as such.

Consider:

casestovars
 /id=Region
 /index=Year
 /fixed=Region_Name.

Region_Name would not be duplicated for each index level.

Transforming Wide to Long

The VARSTOCASES command transforms data from wide to long format.

CategoryRicottone