PSPP-BUG: [bug #57722] Wrong results of NPAR TEST MCNEMAR for some data

[anonymous]

URL:
  <https://savannah.gnu.org/bugs/?57722>

                 Summary: Wrong results of NPAR TEST MCNEMAR for some data
                 Project: PSPP
            Submitted by: None
            Submitted on: Sun 02 Feb 2020 08:41:37 AM UTC
                Category: Numerical Errors
                Severity: 5 - Average
                  Status: None
             Assigned to: None
             Open/Closed: Open
         Discussion Lock: Any
                 Release: None
                  Effort: 0.00

    _______________________________________________________

Details:

The wrong results are the same in the two PSPP versions I have tested:
  * 1.2.0-3 on Debian Buster 64-bit;
  * pspp-20200117-daily-32bits-ForTestingOnly on Windows 7 64-bit.

Steps to reproduce:
1. Open the attached files mcnemar-bug.sav and mcnemar-bug.sps in PSPP.
2. Run all of mcnemar-bug.sps.

Actual results:

NPAR TEST
        /MCNEMAR var1 WITH  var2 (PAIRED).

var1 & var2
╔════╦═════╗
║var1║ var2║
║    ╟──┬──╢
║    ║0 │1 ║
╠════╬══╪══╣
║0   ║ 9│ 1║
║1   ║ 8│ 5║
╚════╩══╧══╝

Test Statistics
╔═══════════╤══╤═════════════════════╤═════════════════════╤
║           │ N│Exact Sig. (2-tailed)│Exact Sig. (1-tailed)│
╠═══════════╪══╪═════════════════════╪═════════════════════╪
║var1 & var2│23│                1.996│                 .998│
╚═══════════╧══╧═════════════════════╧═════════════════════╧

_(this column didn't fit into the bug report field width)_ Point Probability:
.02


NPAR TEST
        /MCNEMAR var2 WITH  var1 (PAIRED).

var2 & var1
╔════╦═════╗
║var2║ var1║
║    ╟──┬──╢
║    ║0 │1 ║
╠════╬══╪══╣
║0   ║ 9│ 8║
║1   ║ 1│ 5║
╚════╩══╧══╝

Test Statistics
╔═══════════╤══╤═════════════════════╤═════════════════════╤
║           │ N│Exact Sig. (2-tailed)│Exact Sig. (1-tailed)│
╠═══════════╪══╪═════════════════════╪═════════════════════╪
║var2 & var1│23│                 .039│                 .020│
╚═══════════╧══╧═════════════════════╧═════════════════════╧

_(this column didn't fit into the bug report field width)_ Point Probability:
.02

Expected results:
The second result may be correct - it is the same as the result of a sign
test.
McNemar test is symmetrical, therefore the results must be identical when the
variables are swapped. However both Exact Sigs differ. Moreover, in the first
result Exact Sig. (2-tailed) is greater than 1, which is impossible because
this is a probability value.



    _______________________________________________________

File Attachments:


-------------------------------------------------------
Date: Sun 02 Feb 2020 08:41:37 AM UTC  Name: mcnemar-bug.sav  Size: 531B   By:
None

<http://savannah.gnu.org/bugs/download.php?file_id=48329>
-------------------------------------------------------
Date: Sun 02 Feb 2020 08:41:37 AM UTC  Name: mcnemar-bug.sps  Size: 93B   By:
None

<http://savannah.gnu.org/bugs/download.php?file_id=48330>

    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?57722>

_______________________________________________
  Message sent via Savannah
  https://savannah.gnu.org/