# t-test on a table

7 messages
Open this post in threaded view
|

## 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       sdTreatment A       25           186         11Treatment 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  ===================== 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
Open this post in threaded view
|

## Re: t-test on a table

 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: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       sdTreatment A       25           186         11Treatment 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  ===================== 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 K Peck[hidden email] ===================== 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
Open this post in threaded view
|

## Re: t-test on a table

Open this post in threaded view
|

## Re: t-test on a table

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.

 Report days sample Mean N Std. Deviation 1 179.76 25 11.181 2 230.60 25 13.435 Total 205.18 50 28.443

then:

t-test groups sample (1,2)

/MISSING=LISTWISE

/VARIABLES=days

/CRITERIA=CI(.90).

From: [hidden email] <[hidden email]>
Sent: 02 May 2020 13:01
Subject: t-test again

As close as I can get:

 Report days sample Mean N Std. Deviation 1 179.76 25 11.181 2 230.60 25 13.435 Total 205.18 50 28.443

 Group Statistics sample N Mean Std. Deviation Std. Error Mean days 1 25 179.76 11.181 2.236 2 25 230.60 13.435 2.687

0.90

 Independent Samples Test Levene's Test for Equality of Variances t-test for Equality of Means F Sig. t df Sig. (2-tailed) Mean Difference Std. Error Difference 90% Confidence Interval of the Difference Lower Upper days Equal variances assumed 1.674 .202 -14.543 48 .000 -50.840 3.496 -56.703 -44.977 Equal variances not assumed -14.543 46.468 .000 -50.840 3.496 -56.707 -44.973

0.95

 Independent Samples Test Levene's Test for Equality of Variances t-test for Equality of Means F Sig. t df Sig. (2-tailed) Mean Difference Std. Error Difference 95% Confidence Interval of the Difference Lower Upper days Equal variances assumed 1.674 .202 -14.543 48 .000 -50.840 3.496 -57.869 -43.811 Equal variances not assumed -14.543 46.468 .000 -50.840 3.496 -57.875 -43.805

John F Hall MA (Cantab) Dip Ed (Dunelm)

Email: [hidden email]

Website: Journeys in Survey Research

-----Original Message-----
From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Bruce Weaver
Sent: 02 May 2020 19:45
To: [hidden email]
Subject: Re: t-test on a table

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)

>

>

>

> Email:

> johnfhall@

>  &lt;mailto:

> johnfhall@

> &gt;

>

> Website: Journeys in Survey Research

> &lt;https://surveyresearch.weebly.com/&gt;

>

> Course: Survey Analysis Workshop (SPSS)

> &lt;https://surveyresearch.weebly.com/1-survey-analysis-workshop-spss.

> html&gt;

>

>

>

> =====================

> 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

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.

--

=====================

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
Open this post in threaded view
|

## Re: t-test on a table

 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   ===================== 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