How to obtain critical values for comparing column proportions?

[Bre]

I conducted a 2 x 3 Chi-square test of homogeneity. I selected to compare the column proportions and selected to adjust p-values using the Bonferroni correction. I also requested the adjusted standardized residuals.

I am wondering what the critical value was the program used to compare columns and determine significant differences. This information is not provided in the output.

Any help would be greatly appreciated. Thanks!

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[David Marso-2]

Ask yourself (or Google) WTF is a "Bonferroni correction"?
How many possible comparisons in your situation.
Do the ma
th.  Or should we do that for you too?

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Jon Peck]

 Assuming you have at least V24, you can get the adjusted sig levels (to three figures) by specifying MERGE=NO SHOWSIG=YES on the /COMPARETEST subcommand or check the corresponding boxes in the Test panel of the CTABLES UI.

On Sun, Jun 9, 2019 at 3:13 AM Bre <[hidden email]> wrote:

I conducted a 2 x 3 Chi-square test of homogeneity. I selected to compare the column proportions and selected to adjust p-values using the Bonferroni correction. I also requested the adjusted standardized residuals.

I am wondering what the critical value was the program used to compare columns and determine significant differences. This information is not provided in the output.

Any help would be greatly appreciated. Thanks!

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

--
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

As far as I remember, Bonferroni divides your alpha (e.g. 5%) by the number of comparisons to obtain the alpha per comparison (familywise error rate). This is more conservative and allows less power than Benjamini-Hochberg which can also be chosen in the CTABLES syntax mentioned by Jon. BH builds a ranking of your p values and uses alphas per comparisons that depend on the position within that ranking.

Am Sonntag, 9. Juni 2019, 19:13:20 MESZ hat Jon Peck <[hidden email]> Folgendes geschrieben:

 Assuming you have at least V24, you can get the adjusted sig levels (to three figures) by specifying MERGE=NO SHOWSIG=YES on the /COMPARETEST subcommand or check the corresponding boxes in the Test panel of the CTABLES UI.

On Sun, Jun 9, 2019 at 3:13 AM Bre <[hidden email]> wrote:

I conducted a 2 x 3 Chi-square test of homogeneity. I selected to compare the column proportions and selected to adjust p-values using the Bonferroni correction. I also requested the adjusted standardized residuals.

I am wondering what the critical value was the program used to compare columns and determine significant differences. This information is not provided in the output.

Any help would be greatly appreciated. Thanks!

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

--
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Jon Peck]

Thanks, Mario.

Quoting a Wikipedia article,

"The false discovery rate (FDR) is a method of conceptualizing the rate of type I errors in null hypothesis testing when conducting multiple comparisons. FDR-controlling procedures are designed to control the expected proportion of "discoveries" (rejected null hypotheses) that are false (incorrect rejections). FDR-controlling procedures provide less stringent control of Type I errors compared to familywise error rate (FWER) controlling procedures (such as the Bonferroni correction), which control the probability of at least one Type I error. Thus, FDR-controlling procedures have greater power, at the cost of increased numbers of Type I errors."  

BH controls the FDR.  The article can be found here for more details:

https://en.wikipedia.org/wiki/False_discovery_rate  

The STATS PADJUST extension command provides six methods for adjusting sig levels for multiple comparisons, but only Bonferroni and BH are provided in CTABLES.  That seemed like plenty for CTABLES purposes, but the unadjusted sig levels could be collected from CTABLES and used with other adjustments.

On Mon, Jun 10, 2019 at 4:39 PM Mario Giesel <[hidden email]> wrote:

As far as I remember, Bonferroni divides your alpha (e.g. 5%) by the number of comparisons to obtain the alpha per comparison (familywise error rate). This is more conservative and allows less power than Benjamini-Hochberg which can also be chosen in the CTABLES syntax mentioned by Jon. BH builds a ranking of your p values and uses alphas per comparisons that depend on the position within that ranking.

Am Sonntag, 9. Juni 2019, 19:13:20 MESZ hat Jon Peck <[hidden email]> Folgendes geschrieben:

 Assuming you have at least V24, you can get the adjusted sig levels (to three figures) by specifying MERGE=NO SHOWSIG=YES on the /COMPARETEST subcommand or check the corresponding boxes in the Test panel of the CTABLES UI.

On Sun, Jun 9, 2019 at 3:13 AM Bre <[hidden email]> wrote:

I conducted a 2 x 3 Chi-square test of homogeneity. I selected to compare the column proportions and selected to adjust p-values using the Bonferroni correction. I also requested the adjusted standardized residuals.

I am wondering what the critical value was the program used to compare columns and determine significant differences. This information is not provided in the output.

Any help would be greatly appreciated. Thanks!

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

--
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

--
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD