Replace not replacing

[Michael Sharp]

Dear Folks,

Using Stata 15.1.

When replacing, Stata is declaring that 134 real changes are made, however when I browse or tab the changes, Stata declared "no observations" (see below, for example). I'm a little lost here. Thanks in advance for your help. Code as follows:

. replace coicop10b=111005099 if inlist(coicop10,111005001,111005003,111005005,1110 05006,111005098,111005099)
(134 real changes made)

. tab interview__key if coicop10b==111005099
no observations

Best,
Mike

[Nick Cox]

replace did replace, but not as you wanted.

This is almost certainly a precision problem. You don't give a data example (FAQ Advice #12) but the problem can be guessed from output like this:

. clear

. set obs 1
number of observations (_N) was 0, now 1

. gen foo = 111005099

. format %9.0f foo

. list

     +-----------+
     |       foo |
     |-----------|
  1. | 111005096 |
     +-----------+

. gen long bar = 111005099

. format %9.0f bar

. list

     +-----------------------+
     |       foo         bar |
     |-----------------------|
  1. | 111005096   111005099 |
     +-----------------------+

There aren't enough bits in your variable, presumably a float, to hold large integers of that magnitude exactly. You need a long type for coicop10b

[Michael Sharp]

SOLVED

Dear Folks, I've sorted this out.

variable coicop10 is a double while variable coicop10b is a float. I solved by: recast double coicop10.

Thank you.

Best,
Mike

[Nick Cox]

#3 No, as recasting a double as a double changes nothing.

Also, recasting a float as a double will not restore detail lost.

You need not only a storage type fit for purpose, but to repeat all calculations producing the new variable.

[Michael Sharp]

Dear Nick

Thank you and indeed, I omitted to state this. I recasted a float as a double before I replaced.

Thanks again,
Mike