INF in SPSS

[Kirill Orlov]

What value could be recommended to use in most practical situations in SPSS to mimic the INF (infinitely large) number, supervalue. SPSS lacks INF, but it is often needed - especially when working in MATRIX session.

[jkpeck]

The largest possible double precision floating point number is about
1.79769 * 10**308
For single precision it would be
3.40282 * 10**38
so any number around that would serve as INF
In real life, you would never come close to that value.
The IEEE 754 standard has all the details.

[Art Kendall]

Many years ago one of the comparisons between SPSS and SAS was that SPSS was all double precision and SAS had a mix of double and single precision.

IIRC SAS changed to all double precision about 30 years ago.  Does SPSS have any single precision arithmetic?

Does MATRIX know about user missing values?  Sysmis?

[jkpeck]

Everything in Statistics is double precision, except, IIRC, the ALSCAL procedure, which was written externally, may have been single precision at one time.  I don't know for sure that it was ever converted.  PREFSCAL is recommended in its place, but ALSCAL remains for backward compatibility.

Missing in MATRIX is handled in the GET command in the procedure.
GET variable reference
[/FILE={file reference}]
            {* }
[/VARIABLES = variable list]
[/NAMES = names vector]
[/MISSING = {ACCEPT}]
                    {OMIT }
                    {value }
[/SYSMIS = {OMIT }]
                  {value}
See the CSR for details.

[Kirill Orlov]

So, Jon, value, say, 1E308 will be admissible to use for Inf purpose in transforms or in Matrix, on all versions of SPSS for the past 10 years, both on a 64bit OS and 32bit OS. Would be that correct to say so?

[jkpeck]

I think so.