Sig testing in a table with a average value
Locked
Sig testing in a table with a average value
 
|
Hi Team,
I have run a table in spss and wanted to sig test it's each cell with a average value at 95% CI. See the attached table :
test.xlsx
I have to sig test each cell of this table with average value 36.61 at 95% level and need to highlighted the cell as below :
Box Red if LOWER than Average
Box Yellow if NO DIFFERENCE than Average
Box GREEN if HIGHER than Average
Please suggest.
Gaurav
|
Re: Sig testing in a table with a average value
Administrator
|
You've not said what the values in the cells of your table are. Given that you want to compare each of them to an "average value 36.61", I guess they are means, and that you are asking how to carry out a whole bunch of single-sample t-tests. Is that right? If it is, you'll need the SD and n for each cell so that you can compute the SE (which is the denominator for the t-test).
But given the number of cells in your table, your family-wise alpha will be very high if you set the per-contrast alpha to .05. (I assume that's what you mean when you said "at 95% CI".) HTH.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
Re: Sig testing in a table with a average value
|
Bruce,
I can imagine a couple of other possibilities, since so little information is given. For instance, what is the shape of the table? Is it one column? Where does the 36.61 come from? Is the 36.61 the mean for *this* table? The usual LSD testing is a test against the overall mean, so the pooled error is not ruled out. If the 36.61 is the mean for a control group, which is a part of the table, or could be, then you not only may use the pooled error, but you have a less stringent correction than Bonferroni (if you want corrections). -- Rich Ulrich > Date: Sun, 15 Jul 2012 15:55:34 -0700 > From: [hidden email] > Subject: Re: Sig testing in a table with a average value > To: [hidden email] > > You've not said what the values in the cells of your table are. Given that > you want to compare each of them to an "average value 36.61", I guess they > are means, and that you are asking how to carry out a whole bunch of > single-sample t-tests. Is that right? If it is, you'll need the SD and n > for each cell so that you can compute the SE (which is the denominator for > the t-test). > > But given the number of cells in your table, your family-wise alpha will be > very high if you set the per-contrast alpha to .05. (I assume that's what > you mean when you said "at 95% CI".) > > HTH. > > > > GauravSrivastava wrote > > > > Hi Team, > > > > I have run a table in spss and wanted to sig test it's each cell with a > > average value at 95% CI. See the attached table : > > http://spssx-discussion.1045642.n5.nabble.com/file/n5714234/test.xlsx > > test.xlsx > > I have to sig test each cell of this table with average value 36.61 at 95% > > level and need to highlighted the cell as below : > > > > Box Red if LOWER than Average > > Box Yellow if NO DIFFERENCE than Average > > Box GREEN if HIGHER than Average > > > > Please suggest. > > > > Gaurav |
Re: Sig testing in a table with a average value
|
|
Hi Rich, Bruce,
Yes, you are correct. 36.61 is the mean of my main table which has all properties ever tested. and this table has the properties only for this round.
This is a table for %age of top4 box for each properties. Please let me know if there is any way to do this either in spss or excel. Gaurav
On Mon, Jul 16, 2012 at 4:53 AM, Rich Ulrich <[hidden email]> wrote:
|
Re: Sig testing in a table with a average value
|
Administrator
|
In reply to this post by Rich Ulrich
The table is 10 rows x 9 columns, so 90 contrasts. That's what had me alarmed about the family-wise alpha.
Judging from the OP's follow-up, there have been multiple "rounds" of data collection. The current 10 x 9 table has data from the latest round of data collection. The 36.61 is the grand mean of *all* data collected thus far -- including the 10 x 9 table, presumably. Comparing subset to whole is not conventional. For independence, the grand mean must exclude the subset. I can think of two ways to achieve that: 1) use grand mean from all prior rounds; or 2) use a grand mean that excludes the current cell (so a different grand mean for every cell). For option 1, it could be done via Dunnett's test (to control family-wise alpha), but with 90 contrasts, good luck. I think the OP needs to think about how to whittle those 90 contrasts down to a smaller number of meaningful contrasts. HTH.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
Re: Sig testing in a table with a average value
|
|
This question really breaks down into two
parts: "could you" and "should you". The "should
you" is pretty dubious given the number of tests. Would the
user be satisfied with say, a Bonferroni-corrected significance level reported
that would be much larger than usual (even assuming that the SDs are available)?
Or perhaps it would be better to abandon an attempt at a true significance
level and just pick a difference that is substantively interesting and
look for suggestive patterns in the table.
The "could you" question, then comes down to whether it is possible to highlight the cells in something like the described manner. That could be done pretty easily with a little scripting in SPSS. The easiest way would be to use the SPSSINC MODIFY TABLES command with a small scriptlet plugin to apply the background coloring according to whatever rule is adopted. It could also be done with VBA scripting in Excel. Jon Peck (no "h") aka Kim Senior Software Engineer, IBM [hidden email] new phone: 720-342-5621 From: Bruce Weaver <[hidden email]> To: [hidden email] Date: 07/16/2012 07:00 AM Subject: Re: [SPSSX-L] Sig testing in a table with a average value Sent by: "SPSSX(r) Discussion" <[hidden email]> The table is 10 rows x 9 columns, so 90 contrasts. That's what had me alarmed about the family-wise alpha. Judging from the OP's follow-up, there have been multiple "rounds" of data collection. The current 10 x 9 table has data from the latest round of data collection. The 36.61 is the grand mean of *all* data collected thus far -- including the 10 x 9 table, presumably. Comparing subset to whole is not conventional. For independence, the grand mean must exclude the subset. I can think of two ways to achieve that: 1) use grand mean from all prior rounds; or 2) use a grand mean that excludes the current cell (so a different grand mean for every cell). For option 1, it could be done via Dunnett's test (to control family-wise alpha), but with 90 contrasts, good luck. I think the OP needs to think about how to whittle those 90 contrasts down to a smaller number of meaningful contrasts. HTH. Rich Ulrich-2 wrote > > Bruce, > I can imagine a couple of other possibilities, since so little > information is given. For instance, what is the shape of the > table? Is it one column? Where does the 36.61 come from? > > Is the 36.61 the mean for *this* table? > The usual LSD testing is a test against the overall mean, > so the pooled error is not ruled out. > > If the 36.61 is the mean for a control group, which is a part > of the table, or could be, then you not only may use the pooled > error, but you have a less stringent correction than Bonferroni > (if you want corrections). > > -- > Rich Ulrich > > >> Date: Sun, 15 Jul 2012 15:55:34 -0700 >> From: bruce.weaver@ >> Subject: Re: Sig testing in a table with a average value >> To: SPSSX-L@.UGA >> >> You've not said what the values in the cells of your table are. Given >> that >> you want to compare each of them to an "average value 36.61", I guess >> they >> are means, and that you are asking how to carry out a whole bunch of >> single-sample t-tests. Is that right? If it is, you'll need the SD and >> n >> for each cell so that you can compute the SE (which is the denominator >> for >> the t-test). >> >> But given the number of cells in your table, your family-wise alpha will >> be >> very high if you set the per-contrast alpha to .05. (I assume that's >> what >> you mean when you said "at 95% CI".) >> >> HTH. >> >> >> >> GauravSrivastava wrote >> > >> > Hi Team, >> > >> > I have run a table in spss and wanted to sig test it's each cell with a >> > average value at 95% CI. See the attached table : >> > http://spssx-discussion.1045642.n5.nabble.com/file/n5714234/test.xlsx >> > test.xlsx >> > I have to sig test each cell of this table with average value 36.61 at >> 95% >> > level and need to highlighted the cell as below : >> > >> > Box Red if LOWER than Average >> > Box Yellow if NO DIFFERENCE than Average >> > Box GREEN if HIGHER than Average >> > >> > Please suggest. >> > >> > Gaurav > ... > ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Sig-testing-in-a-table-with-a-average-value-tp5714234p5714244.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |