Once Written, Twice Applied: The Basics of Array Processing In SAS Elizabeth A. Roth, RAND Corporation, Santa Monica, CA

Transcription

1 Once Written, Twice Applied: The Basics of Array Processing In SAS Elizabeth A. Roth, RAND Corporation, Santa Monica, CA ABSTRACT Using arrays in DATA step programming can streamline programming in important ways. This paper introduces novice users to the basics of array processing through examples and explanations of syntax. In addition to the array statement, do-loop processing and the DIM function are discussed. Finally, several specific uses for arrays are described to help users get started programming with arrays. INTRODUCTION The power and flexibility of SAS DATA step programming is uncovered when one can reduce the amount of code written without sacrificing functionality, processing time, or readability. All programmers can appreciate the convenience of having fewer lines of code to write, document, and support. Arrays are one of the many tools SAS provides that allow programmers to streamline their code. The novice programmer who takes the time to add arrays to their kit of DATA step tools will be rewarded with elegant code that is easy to read, develop, and maintain. AN EXAMPLE The usefulness of arrays in DATA step programming is easily illustrated with an example. Let s say we have been asked to clean some data from a survey conducted by the local zoo, trying to investigate which animals patrons visited. Here is the survey question they might have administered: (1) Which animals did you see during your visit to the zoo today? Please check all that apply: Monkeys Giraffes Hippos Elephants Tapirs Zebras Orangutans None of the above Your data entry provider entered 12,000 surveys coding the checked boxes as 1 and the unchecked boxes as 2. You would like to store the unchecked boxes as zeros so you can compute means for each of the variables. Here is an example of how you might use the DATA step, without arrays, to recode the 2 s to 0 s: data animals; set zoosurvey ; /* Recode 2s to 0s */ if monkeys=2 then monkeys=0 ; if giraffes=2 then giraffes=0 ; if hippos=2 then hippos=0 ; if elephants=2 then elephants=0 ; if tapirs=2 then tapirs=0 ; if zebras=2 then zebras=0 ; if orangutans=2 then orangutans=0 ; run; This is a perfectly functional SAS program that will accomplish your recoding goal. It has the added benefit of being very clear in what is being recoded, however it is inflexible when you want to change how you are recoding every variable (imagine if you wanted to do a more complicated recoding procedure that required several lines of code per variable!) and it requires particularly good keyboarding skills (imagine you needed to recode the variables for each of the hundreds of animals in the zoo!). Arrays allow you to execute the exact same code for many different variables here is an example of how an array might simplify (and beautify!) the same code: 1

2 DATA animals; SET zoosurvey ; ARRAY animals {7} monkeys giraffes hippos elephants tapirs zebras orangutans ; /* Loop through all elements of the array */ DO i=1 TO 7 ; IF animals[i] = 2 THEN animals[i] = 0 ; RUN; This code is much more elegant: it uses fewer lines and requires much less repetition. THE MECHANICS OF THE ARRAY STATEMENT As seen in the example above, there are two parts to working with array statements: defining an array, and using it. In the example above, we used an array in the context of a do loop, referencing each element of the array from start to finish. In this section we ll talk about the specifics of defining and using, or referencing, arrays. There are at least four components to the array statement: the keyword ARRAY, a name that you define for the array, the dimension of the array, and the array elements themselves. Of course like every SAS statement, the ARRAY statement ends with a semicolon. We saw how to define an array for the zoo survey question above; here is another example of an array statement: ARRAY days {7} Monday Tuesday Wednesday Thursday Friday Saturday Sunday ; This statement creates an array called days that contains seven variables, or elements, which are the variables called Monday, Tuesday, Wednesday, etc. The number in curly brackets 1, {7}, is the dimension; this tells SAS how many elements will make up the array. Finally, the statement ends with a list of the variables that are the array s elements, followed by the semi-colon. SAS expects to find exactly the same number of variables listed as defined in the dimension. SAS also insists on finding variables of all the same type, either numeric or character, in the array definition. Implicit in the definition of an array is the fact that the variables are ordered in such a way that they can be referenced numerically, that is to say the first variable listed after the dimension (in this case Monday) is the first variable, the next one ( Tuesday ) is the second variable, the one following that ( Wednesday ) is the third variable, and so on. A variable s ordinal position in the array is called its index. So in this case, the index for the variable Thursday is 4, the index for Friday is 5, Saturday s is 6, and so on. A variable s index in an array is only relevant in the context of the array so if a variable appears in, say three arrays, the same variable will have three indices, one in the context of each array. The index is an important concept whenever you want to do anything with arrays. An element of the array is referenced by the array name plus its index, surrounded by brackets 2, like this: days[3] = 7 ; This code assigns the value seven to the Wednesday variable. Let s look at some code from our zoo example: ARRAY animals {7} monkeys giraffes hippos elephants tapirs zebras orangutans ; animals[2] = 0 ; This code recodes the giraffes variable (the second variable in the animals array) to zero. The index of the array (in the previous example it was 2) can be expressed as a digit (the number 2) or as a variable name, in which case SAS will substitute the current value of the variable. For example, we could create a variable 1 The dimension can be surrounded by parentheses, curly brackets, or straight brackets. I prefer to use curly brackets so as to avoid confusion with other SAS notation, such as function parameters. 2 Like the dimension, the index can be surrounded by parentheses, curly brackets, or straight brackets, but to avoid confusion I prefer to use straight brackets only. 2

3 whose value equals 2 and substitute the variable name as the index in the call to the array: /* Create a constant */ myvar = 2 ; /* Now use the constant */ animals[myvar] = 0 ; This code does exactly the same thing as the previous example: recodes the giraffes variable to zero. The ability to refer to an array s index by a variable name is the key to the power and flexibility of arrays. You can use SAS DO loops 3 to change the value of the index variable automatically, thus executing the same code on each variable in the array. This technique is probably the most common way to see calls to arrays. Let s look at another simple example from our days of the week array: /* Loop through all elements of the array */ DO myloopvar = 1 TO 7 ; IF days[myloopvar] = 1 THEN days[myloopvar] = 0 ; By incrementing the index variable from the starting value (1) through all values up until the stopping value (7) you execute the IF days[myloopvar] = 1 THEN days[myloopvar] = 0 line exactly seven times, once for each value of myloopvar, or in other words once for each element, or variable, in the array. Executing this code is equivalent to executing the following code: IF days[1] = 1 THEN days[1] = 0 ; IF days[2] = 1 THEN days[2] = 0 ; IF days[3] = 1 THEN days[3] = 0 ; IF days[4] = 1 THEN days[4] = 0 ; IF days[5] = 1 THEN days[5] = 0 ; IF days[6] = 1 THEN days[6] = 0 ; IF days[7] = 1 THEN days[7] = 0 ; which is exactly the same as executing this code: IF Monday = 1 THEN Monday = 0 ; IF Tuesday = 1 THEN Tuesday = 0 ; IF Wednesday = 1 THEN Wednesday = 0 ; IF Thursday = 1 THEN Thursday = 0 ; IF Friday = 1 THEN Friday = 0 ; IF Saturday = 1 THEN Saturday = 0 ; IF Sunday = 1 THEN Sunday = 0 ; only the version of the code using arrays is much more elegant. There is one last step we can take towards improving our ARRAY statement, which is to allow SAS to compute the stopping value of the DO loop, which is the dimension of the array, automatically. The SAS dim function does this, taking the name of an array as its only parameter and returning the array s dimension. Call the DIM function directly from the do loop and our elegant code becomes: do myloopvar = 1 to dim(days) ; if days[myloopvar] = 1 then days[myloopvar] = 0 ; end; Using the DIM function is useful in cases where the dimension of the array may change from time to time, as in the case with the zoo survey. Should we add 100 new animals to the list, we only need add them to the ARRAY statement and all of our looping code will compute the total number of items in the array at compile-time. However, if you know that the total number of elements in the array is relatively stable, it is preferable to simply enter the dimension as a constant. While there is probably a miniscule processing-time advantage to not having SAS 3 For the uninitiated, DO loops execute the same lines of code over and over again either a given number of times or until a certain condition has been met. The general form of the DO loop is: DO loopvar = 1 TO nloops ; code to be repeated ; 3

4 compute the dimension, it sometimes can be easier to debug array code when the dimension is entered as a constant. USES FOR ARRAYS You have seen how arrays can be powerful and yet simple to program. Let s look at a few uses for arrays that exploit their full capacity. This section demonstrates additional ways to define, call, or use arrays that just may solidify their place as one of your favorite tools in your programming toolbox. SETTING THE VALUES OF AN ARRAY S INDICES An array s index need not always begin at one. Indeed, you can define the dimension of an array in such as way that the indices are actually descriptive of the elements in the array. Say for example you conduct the Zoo survey in three consecutive months, May, June, and July, and you have entered the number of respondents in May, June, and July in a separate dataset [SHOW DATASET]. If you put these three variables into an array, it would make sense to have the index correspond to the month of the year. Simply defining the array as follows will accomplish exactly this: ARRAY nresp {5:7} may june july ; USING INDICES THAT MEAN SOMETHING The index of the variable may in the nresp array is 5, as opposed to 1. Now say you have a new set of observations and you d like to tally the months of their responses based on the survey date, survdate. This line of code will accomplish exactly that: nresp [ month(survdate) ] + 1 ; REFERENCING THE INDEX WITH A VARIABLE NAME, PART 2 In one of the above examples, we referenced a variable in the days array using a variable name for the index in the array call, as opposed to a constant ( animals[myvar] as opposed to animals[2], where myvar was set equal to 2). Naming the index variable cleverly can improve readability of the program substantially. Let s look, for example, at some code to top-code thirty years worth of respondents annual reports of their income: ARRAY income {1978:2007} income income2007 ; /* Loop over all elements of the array */ DO survyear = 1978 TO 2007 ; IF income[survyear] gt THEN income[survyear] = ; END ; Notice how the income array was defined using indices that correspond to the years the questions were asked (1978 through 2007), and the starting and stopping values for the do-loop correspond to those years 4. Now whenever you reference an element in the income array, you can use a meaningful index instead of an arbitrary one. This simplification can be particularly relevant when working with longitudinal or repeated-measures data, which can be particularly complex to process. ARRAYS AS FUNCTION PARAMETERS Once defined, you can use an array name to refer to all of the variables in an array at once, inside a function: mostsurveys = MAX (of nresp[*]) ; This code computes the max of the three variables, may, june, and july and stores it in the variable called mostsurveys. QUICKLY DEFINING ARRAYS FOR BLOCKS OF VARIABLES If you want to perform the same operation on all of the variables in a dataset, you can use the SAS reserved variable list names _NUMERIC_, _CHARACTER_, or _ALL_ to define the elements of an array: ARRAY numvars {*} _NUMERIC_ ; ARRAY charvars {*} _CHARVARS_ ; 4 As an aside, the DO-loop values only need to directly correspond to the dimension if you want to process all the variables in the array. If you want to process a subset of the variables in the array, simply process the indices you are interested in. 4

5 ARRAY allvars {*} _ALL_ ; Note, if you use _ALL_, all the variables in the dataset must have previously been defined as all the same data type. This application is the perfect example of when to allow SAS to compute the number of elements in the array on the fly, and when to use the DIM function in the looping code. CONCLUSION SAS arrays are an important basic programming tool that beginning users should take the time to acquire. Combined with do looping, arrays allow users to process multiple variables with the same code. Arrays can be passed to many SAS functions as a quick way to get information about the array, or the variables within the array. And there are several shortcuts that programmers can use to quickly define large blocks of variables as an array. Once learned, the ARRAY statement will be a tool that will serve to make your programs much more readable, effective, and generally more elegant. ACKNOWLEDGMENTS Thanks to Lara Hilton for the invitation to present and for the thoughtful read of this paper. CONTACT INFORMATION Your comments and questions are valued and encouraged. Contact the author at: Elizabeth Roth RAND Corporation 1776 Main Street Santa Monica, CA (310) Elizabeth_Roth@rand.org SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. indicates USA registration. Other brand and product names are trademarks of their respective companies. 5