TRANSFORM <data> BY <function of X>
transforms all the active variables and observations (possibly limited
by MASK, IND and CASES specifications) by <function of X> where X
refers to one of the active variables in turn.
TRANSFORM accepts the same functions as VAR. Thus functions
sqrt,log,exp,sin,cos,tan,atn,abs,int,probit, and rnd are available.
The 'if' structure is allowed:
For example,
TRANSFORM DATA1 BY if(X=9)then(MISSING)else(X)
replaces all values '9' in active variables by missing values.
Library functions and the control statement 'for' are not accepted.
Although the same transformations can be made by VAR for each variable
separately, use of TRANSFORM saves time and effort in systematic modi-
fications of data values.
For example,
TRANSFORM YEAR88 BY log(X)
replaces data values in data YEAR88 by their natural logarithms.
The transformed values can be saved as other (possibly new) variables
by using TRANSFORM in an extended form
TRANSFORM <data> BY <function of X> AS <letter>:<type>
where <letter> is a character to be placed in front of each transformed
variable. The extension :<type> where <type> is 1,2,4 or 8 is optional
and gives the type of new numerical variables. Default is 4.
For example,
TRANSFORM YEAR88 BY log(X) AS L
VARS=Sales,Costs
makes the logarithms of variables Sales and Costs as (new) variables
LSales and LCosts.
Also extra variables may be used in the transformation.
In order to speed up the process, in such cases a specification MODE=2
should be entered.
Example: "Proportions of events of Decathlon in percentages"
>COPY <Survo>\U\D\DECA.SVO TEST.SVO
MASK=--AAAAAAAAAA--
MODE=2
TRANSFORM TEST BY 1000*X/Points
FILE SHOW TEST
The active variables and observations can be centered and standardised
by
TRANSFORM <data> BY CENTERING
and
TRANSFORM <data> BY STANDARDISING (or STD),
respectively.
TRANSFORM <data> BY #UNIFORM(a,b)
generates values from a uniform distribution on interval (a,b).
For continuos variables (types 4,8) values will be from a continuos
uniform distribution on (a,b).
For discrete variables (types 1,2) values will be from a discrete
uniform distribution with values a, a+1, a+2,..., b-1 .
In huge datasets this alternative is many times faster than
TRANSFORM <data> BY (b-a)*rand(seed)+a
The pseudo-random number generator and its seed number can be selected by
a RND specification. Example: RND=rand(1998115)
TRANSFORM <data> BY #LINEAR(a,b) makes linear transformation X -> a*X+b .
TRANSFORM <data> BY #DISTR(P)
makes random values from a discrete distribution.
The values (in ascending order) and probabilities
are given as a matrix file P of two columns.
In all applications of TRANSFORM where variables are transformed
`in place' it is worthwhile to notice that transformations leading
to decimal values (like log(X)) may be inaccurate in case of integer-
valued variables (of type 1 or 2). In such a situation one should
take a list of all active variables into the edit field by FILE STATUS,
change types 1 and 2 to either 4 or 8, create (FILE CREATE) a new data
file with this list of variables and copy (FILE COPY) the original data
to this new data file. Transformations can then be made in the new data
file.
M = Simulation of Markov chains by TRANSFORM
V = More information on transformations
Help System (web edition)