inverse error function

[paulvonhippel]

Are there any implementations of the inverse error function for Stata?

[daniel klein]

Does this help?

[paulvonhippel]

Thanks, daniel klein . The earlier thread discussed the error function, while I want the inverse error function. But the basic strategy is similar. I can just transform the results of the invnormal() function. For example, here is the inverse normal function of 0.2:

display invnormal((.2 + 1) / 2) / sqrt(2)

Bonus round: is is possible to define a new function that calculates the inverse normal function for any argument (not just 0.2).

[daniel klein]

Bonus round: is is possible to define a new function that calculates the inverse normal function for any argument (not just 0.2).

No. You cannot implement new Stata functions. It's straightforward in Mata, though:

. mata : real matrix inverror(real matrix x) return( invnormal((x:+1)/2)/sqrt(2) )

. mata : inverror((.2,.42,.73))
                 1             2             3
    +-------------------------------------------+
  1 |  .1791434546   .3913020878   .7799830136  |
    +-------------------------------------------+

[Nick Cox]

What you can also do is map a variable to a variable:

gen wanted = invnormal((x + 1) / 2) / sqrt(2)