topic Re: Doing a sum on a text field in SAS Programming

Doing a sum on a text field

zdassu — Wed, 01 Nov 2017 12:11:58 GMT

Hi I am trying to sum a column in a table, how ever although the column contains numbers they have been saved as text. I know in SQL you can use the following code but I am struggling to use this code in SAS

select SUM(cast(AAAAAAAAAA as float))/1000000 from ##Temp1

Your help in this matter is much appreciated

Re: Doing a sum on a text field

PeterClemmensen — Wed, 01 Nov 2017 12:14:09 GMT

What does your data look like?

Sounds like you need to convert the character values to numeric first and then calculate the sum from there, eg with PROC MEANS.

Re: Doing a sum on a text field

zdassu — Wed, 01 Nov 2017 12:17:46 GMT

The data looks like this, quite a few decimal places

7333.83706005760236
-241.48477505034183
-3721.90979470953705
282.78293262436968
8401.59702862381179
319.66827473169031
9488.22997865952625
276.11825793635553
8379.37965250339576
1001.47034302957246

Re: Doing a sum on a text field

RW9 — Wed, 01 Nov 2017 12:19:21 GMT

The equivalent of cast in SAS is input/put. Put takes a numeric and creates a character based on the format, input takes a character and returns a numeric based on the informat. Now these will give you a warning if there are items that cannot be converted, e.g. ABC cannot be numeric, these will be missing which is dot. So yu could do: select sum(input(text_var,best.)) as result

Re: Doing a sum on a text field

zdassu — Wed, 01 Nov 2017 12:27:31 GMT

Thanks for your help, worked just as you described

Re: Doing a sum on a text field

ballardw — Wed, 01 Nov 2017 13:57:41 GMT

With that many decimals you may run into numeric precision issues.

Re: Doing a sum on a text field

zdassu — Wed, 01 Nov 2017 14:24:56 GMT

DO you have any other suggestions? I don't mind it rounding the decimals up to about 6-7 places

Re: Doing a sum on a text field

ballardw — Wed, 01 Nov 2017 22:54:03 GMT

@zdassu wrote:

DO you have any other suggestions? I don't mind it rounding the decimals up to about 6-7 places

I brought up the issue in case you considered all of those digits as significant. Your calculations involving such numbers could be off in the last one or two decimals (or more if enough calculations are done). You just need to know your data and treat as appropriate for your purpose.

I obviously don't have your data but if there are digits that are not actually significant that you carry through calculations then your result may imply precision that does not exist which has a chance of creating final statistics that are/are not significant based on using those values in a regression or similar procedure.