t-test on a table
Locked
 
|
Is there a way in SPSS of running t-test on, or generating a data set from, an actual table? Fictitious data on cancer survival rates in mean days: Mean N days sd Treatment A 25 186 11 Treatment B 25 203 13 Associated question: In a population with a certain type of cancer, the failure rate of treatment X is 15%. Researchers wish to draw 2 equal size samples; one will receive treatment X the other treatment Y. How large do the two samples need to be to reduce the failure rate to 12% at 0.05 sig and 0.95 confidence? A sample of 1000 receives treatment Z, with a failure rate reduced to 10%: is this significant? John F Hall MA (Cantab) Dip Ed (Dunelm) IBM-SPSS Academic Author 9900074 Email: [hidden email] Website: Journeys in Survey Research Course: Survey Analysis Workshop (SPSS) |
|
CTABLES compare column means test could be used here. Or, if you just have the aggregates, you could use the SPSSINC SUMMARY TTEST extension command, which can be installed from the Extensions > Extension Hub menu. Here is the syntax for your example. SPSSINC SUMMARY TTEST N1=25 MEAN1=286 SD1=22 LABEL1="Treatment A" N2=35 MEAN2=203 SD2=13 LABEL2="Treatment B" CI=95. It does the test with and without the equal variance assumption and provides asymptotic and exact confidence intervals. Spoiler alert: in this example the difference is highly significant. On Sat, May 2, 2020 at 2:09 AM Joihn F Hall <[hidden email]> wrote:
|
Administrator
|
In reply to this post by John F Hall
Re the first question, use ONEWAY with matrix input and compute the square
root of F if you need to report it as a t-test. * Oneway ANOVA using summary data. DATA LIST LIST / ROWTYPE_ (a8) grp (f5.0) VARNAME_ (a15) Y (f8.2) . BEGIN DATA "MEAN" 1 "Treatment A" 186 "STDDEV" 1 "Treatment A" 11 "N" 1 "Treatment A" 25 "MEAN" 2 "Treatment B" 203 "STDDEV" 2 "Treatment B" 13 "N" 2 "Treatment B" 25 END DATA. ONEWAY Y BY grp / MATRIX = in(*) / STATISTCS = DESCRIPTIVES WELCH BROWNFORSYTHE. * If you need to report it as a t-test, t = SQRT(F). I'm not sure I understand the "associated question" you posted. Are you asking what sample size is needed to detect the difference between 15% and 12% (assuming equal sample sizes in the two groups)? If so, I don't have access to the SPSS module for sample size estimation. But using Stata, here's what I get. Change to a fixed font to make the table line up properly. . power twoproportions 0.15 0.12, test(chi2) power(0.8 0.9 0.95) Performing iteration ... Estimated sample sizes for a two-sample proportions test Pearson's chi-squared test Ho: p2 = p1 versus Ha: p2 != p1 +-----------------------------------------------------------------+ | alpha power N N1 N2 delta p1 p2 | |-----------------------------------------------------------------| | .05 .8 4,072 2,036 2,036 -.03 .15 .12 | | .05 .9 5,450 2,725 2,725 -.03 .15 .12 | | .05 .95 6,740 3,370 3,370 -.03 .15 .12 | +-----------------------------------------------------------------+ The final question about treatment Z is also unclear. What do you want to compare to the 10% of 1000? Cheers, Bruce John F Hall wrote > Is there a way in SPSS of running t-test on, or generating a data set > from, > an actual table? > > > > Fictitious data on cancer survival rates in mean days: > > > > Mean > > N days sd > > Treatment A 25 186 11 > > Treatment B 25 203 13 > > > > Associated question: > > In a population with a certain type of cancer, the failure rate of > treatment > X is 15%. Researchers wish to draw 2 equal size samples; one will receive > treatment X the other treatment Y. How large do the two samples need to > be > to reduce the failure rate to 12% at 0.05 sig and 0.95 confidence? > > > > A sample of 1000 receives treatment Z, with a failure rate reduced to 10%: > is this significant? > > > > John F Hall MA (Cantab) Dip Ed (Dunelm) > > IBM-SPSS Academic Author 9900074 > > > > Email: > johnfhall@ > <mailto: > johnfhall@ > > > > Website: Journeys in Survey Research > <https://surveyresearch.weebly.com/> > > Course: Survey Analysis Workshop (SPSS) > <https://surveyresearch.weebly.com/1-survey-analysis-workshop-spss.html> > > > > > ===================== > To manage your subscription to SPSSX-L, send a message to > LISTSERV@.UGA > (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 ----- -- 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. -- Sent from: http://spssx-discussion.1045642.n5.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
--
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/). |
|
|
Bruce’s solution is much better than mine. I did a pretty clumsy workaround. Generated a data set with 50 cases, 25 with sample =1 and 25 with sample = 2. Couldn’t get exact Ns and SDs, but after about 40 attempts got pretty close. if sample = 1 days = TRUNC(RV.UNIFORM(160, 201)). if sample = 2 days = TRUNC(RV.UNIFORM(209, 253)). means days by sample.
then: t-test groups sample (1,2) /MISSING=LISTWISE /VARIABLES=days /CRITERIA=CI(.90). From: [hidden email] <[hidden email]> Maddy As close as I can get:
0.90
0.95
John F Hall MA (Cantab) Dip Ed (Dunelm) IBM-SPSS Academic Author 9900074 Email: [hidden email] Website: Journeys in Survey Research Course: Survey Analysis Workshop (SPSS) -----Original Message----- Re the first question, use ONEWAY with matrix input and compute the square root of F if you need to report it as a t-test. * Oneway ANOVA using summary data. DATA LIST LIST / ROWTYPE_ (a8) grp (f5.0) VARNAME_ (a15) Y (f8.2) . BEGIN DATA "MEAN" 1 "Treatment A" 186 "STDDEV" 1 "Treatment A" 11 "N" 1 "Treatment A" 25 "MEAN" 2 "Treatment B" 203 "STDDEV" 2 "Treatment B" 13 "N" 2 "Treatment B" 25 END DATA. ONEWAY Y BY grp / MATRIX = in(*) / STATISTCS = DESCRIPTIVES WELCH BROWNFORSYTHE. * If you need to report it as a t-test, t = SQRT(F). I'm not sure I understand the "associated question" you posted. Are you asking what sample size is needed to detect the difference between 15% and 12% (assuming equal sample sizes in the two groups)? If so, I don't have access to the SPSS module for sample size estimation. But using Stata, here's what I get. Change to a fixed font to make the table line up properly. . power twoproportions 0.15 0.12, test(chi2) power(0.8 0.9 0.95) Performing iteration ... Estimated sample sizes for a two-sample proportions test Pearson's chi-squared test Ho: p2 = p1 versus Ha: p2 != p1 +-----------------------------------------------------------------+ | alpha power N N1 N2 delta p1 p2 | |-----------------------------------------------------------------| | .05 .8 4,072 2,036 2,036 -.03 .15 .12 | | .05 .9 5,450 2,725 2,725 -.03 .15 .12 | | .05 .95 6,740 3,370 3,370 -.03 .15 .12 | +-----------------------------------------------------------------+ The final question about treatment Z is also unclear. What do you want to compare to the 10% of 1000? Cheers, Bruce John F Hall wrote > Is there a way in SPSS of running t-test on, or generating a data set > from, an actual table? > > > > Fictitious data on cancer survival rates in mean days: > > > > Mean > > N days sd > > Treatment A 25 186 11 > > Treatment B 25 203 13 > > > > Associated question: > > In a population with a certain type of cancer, the failure rate of > treatment X is 15%. Researchers wish to draw 2 equal size samples; > one will receive treatment X the other treatment Y. How large do the > two samples need to be to reduce the failure rate to 12% at 0.05 sig > and 0.95 confidence? > > > > A sample of 1000 receives treatment Z, with a failure rate reduced to 10%: > is this significant? > > > > John F Hall MA (Cantab) Dip Ed (Dunelm) > > IBM-SPSS Academic Author 9900074 > > > > Email: > johnfhall@ > <mailto: > johnfhall@ > > > > Website: Journeys in Survey Research > <https://surveyresearch.weebly.com/> > > Course: Survey Analysis Workshop (SPSS) > <https://surveyresearch.weebly.com/1-survey-analysis-workshop-spss. > html> > > > > > ===================== > To manage your subscription to SPSSX-L, send a message to > (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 ----- -- Bruce Weaver 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. -- Sent from: http://spssx-discussion.1045642.n5.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 |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
In reply to this post by John F Hall
As to generating variables with specific means and SDs -
Generate the variables with the shape of distribution that
you want. For survival times, negative exponential might
be apt.
If the generating doesn't allow you to specify mean and SD...
Then use Descriptives to z-score by group;
multiple, for each group separately, by its desired SD; and
add on the desired Mean.
Bruce handled the (obscurely worded) question about power.
--
Rich Ulrich
From: SPSSX(r) Discussion <[hidden email]> on behalf of Joihn F Hall <[hidden email]>
Sent: Saturday, May 2, 2020 4:09 AM To: [hidden email] <[hidden email]> Subject: t-test on a table Is there a way in SPSS of running t-test on, or generating a data set from, an actual table?
Fictitious data on cancer survival rates in mean days:
Mean N days sd Treatment A 25 186 11 Treatment B 25 203 13
Associated question: In a population with a certain type of cancer, the failure rate of treatment X is 15%. Researchers wish to draw 2 equal size samples; one will receive treatment X the other treatment Y. How large do the two samples need to be to reduce the failure rate to 12% at 0.05 sig and 0.95 confidence?
A sample of 1000 receives treatment Z, with a failure rate reduced to 10%: is this significant?
John F Hall MA (Cantab) Dip Ed (Dunelm) IBM-SPSS Academic Author 9900074
Email: [hidden email] Website: Journeys in Survey Research Course: Survey Analysis Workshop (SPSS)
|
|
Administrator
|
I think I read too quickly the first time, and didn't notice that the context
is survival times. Rich is right on the money when he says the distributions of survival times will be positively skewed. If you did have the raw data, you'd almost certainly want to use some kind of survival (or time-to-event) analysis. I cannot think of anyway to do that from summary statistics. I suppose that if it was a situation where all subjects experienced the "event", I might consider using quantile regression. SPSS 26 has a new QUANTILE REGRESSION command. If you don't have v26, you can use an R-extension command IIRC. HTH. Rich Ulrich wrote > As to generating variables with specific means and SDs - > Generate the variables with the shape of distribution that > you want. For survival times, negative exponential might > be apt. > > If the generating doesn't allow you to specify mean and SD... > Then use Descriptives to z-score by group; > multiple, for each group separately, by its desired SD; and > add on the desired Mean. > > Bruce handled the (obscurely worded) question about power. > > -- > Rich Ulrich ----- -- 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. -- Sent from: http://spssx-discussion.1045642.n5.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
--
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/). |
|
|
This might be irrelevant, but, anyway, speaking as a suspicious
data analyst:
Pushing the "survival data" aspect a little more - John did say
that these were fictitious data. but the reported SDs are far too
small to be realistic if these were based on any proper survivorships.
A difference of more that 1.3 SD does not take a large N to detect.
=== from John's post Fictitious data on cancer survival rates in mean days:
Mean N days sd Treatment A 25 186 11 Treatment B 25 203 13 === end
I do remember, once upon a time, seeing tiny, unrealistic SDs in a
paper submitted for publication. Looking really closely at the text,
I found out that they were Standard Errors. Multiplied by the square
root of N, they became plausible as SDs.
--
Rich Ulrich
From: SPSSX(r) Discussion <[hidden email]> on behalf of Bruce Weaver <[hidden email]>
Sent: Saturday, May 2, 2020 5:09 PM To: [hidden email] <[hidden email]> Subject: Re: t-test on a table I think I read too quickly the first time, and didn't notice that the context
is survival times. Rich is right on the money when he says the distributions of survival times will be positively skewed. If you did have the raw data, you'd almost certainly want to use some kind of survival (or time-to-event) analysis. I cannot think of anyway to do that from summary statistics. I suppose that if it was a situation where all subjects experienced the "event", I might consider using quantile regression. SPSS 26 has a new QUANTILE REGRESSION command. If you don't have v26, you can use an R-extension command IIRC. HTH. Rich Ulrich wrote > As to generating variables with specific means and SDs - > Generate the variables with the shape of distribution that > you want. For survival times, negative exponential might > be apt. > > If the generating doesn't allow you to specify mean and SD... > Then use Descriptives to z-score by group; > multiple, for each group separately, by its desired SD; and > add on the desired Mean. > > Bruce handled the (obscurely worded) question about power. > > -- > Rich Ulrich ----- -- 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. -- Sent from: http://spssx-discussion.1045642.n5.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 |
| Free forum by Nabble | Edit this page |