topic Re: GLM in Statistical Procedures

GLM

zahidhasandipu — Fri, 11 Sep 2020 03:42:40 GMT

I know this is really a long and complicated result, still, I would really appreciate your help.

The hypothesis is students who use software after college known as condition 1 would perform better than the in-class group known as condition 0. To test this, students were given a math test during the first week of a month, and the same test again during the first week of another month.
How should I know whether to accept/reject the hypothesis and is there a change in the student's scores?

Result:

GLM, Repeated Sunday, August 23, 2020 05:52:05 PM 1
The GLM Procedure

Class Level Information
Class	Levels	Values
Condition	2	0 1

Number of Observations Read	353
Number of Observations Used	337

GLM, Repeated Sunday, August 23, 2020 05:52:05 PM 2
The GLM Procedure
Repeated Measures Analysis of Variance

Repeated Measures Level Information
Dependent Variable	pretest	posttest
Level of testscore	1	2

MANOVA Test Criteria and Exact F Statistics for the Hypothesis of no testscore Effect H = Type III SSCP Matrix for testscore E = Error SSCP Matrix S=1 M=-0.5 N=166.5
Statistic	Value	F Value	Num DF	Den DF	Pr > F
Wilks' Lambda	0.96161567	13.37	1	335	0.0003
Pillai's Trace	0.03838433	13.37	1	335	0.0003
Hotelling-Lawley Trace	0.03991649	13.37	1	335	0.0003
Roy's Greatest Root	0.03991649	13.37	1	335	0.0003

MANOVA Test Criteria and Exact F Statistics for the Hypothesis of no testscoreCondition Effect H = Type III SSCP Matrix for testscoreCondition E = Error SSCP Matrix S=1 M=-0.5 N=166.5
Statistic	Value	F Value	Num DF	Den DF	Pr > F
Wilks' Lambda	0.99435945	1.90	1	335	0.1690
Pillai's Trace	0.00564055	1.90	1	335	0.1690
Hotelling-Lawley Trace	0.00567255	1.90	1	335	0.1690
Roy's Greatest Root	0.00567255	1.90	1	335	0.1690

GLM, Repeated Sunday, August 23, 2020 05:52:05 PM 3
The GLM Procedure
Repeated Measures Analysis of Variance
Tests of Hypotheses for Between Subjects Effects

Source	DF	Type III SS	Mean Square	F Value	Pr > F
Condition	1	418.010493	418.010493	17.48	<.0001
Error	335	8009.641852	23.909379

GLM, Repeated
The GLM Procedure

Sunday, August 23, 2020 05:52:05 PM 4

Repeated Measures Analysis of Variance
Univariate Tests of Hypotheses for Within Subject Effects

Source	DF	Type III SS	Mean Square	F Value	Pr > F
testscore	1	92.151681	92.151681	13.37	0.0003
testscore*Condition	1	13.095717	13.095717	1.90	0.1690
Error(testscore)	335	2308.611583	6.891378

GLM, Repeated Sunday, August 23, 2020 05:52:05 PM 5

GLM, Repeated Sunday, August 23, 2020 05:52:05 PM 6

pretest	posttest	Level of Condition	N
Mean	Std Dev	Mean	Std Dev
0	148	68.3993243	4.73396321	68.8635135	3.92975997
1	189	69.7052910	3.44273418	70.7312169	3.66520864

Re: GLM

PaigeMiller — Fri, 11 Sep 2020 11:37:58 GMT

It always helps to show your code. Obviously, you have used MANOVA here, which does not seem to fit the problem description.

Then you say:

The hypothesis is students who use software after college known as condition 1 would perform better than the in-class group known as condition 0. To test this, students were given a math test during the first week of a month, and the same test again during the first week of another month.

but I'm not sure what you mean. Do you mean that students in condition 1 performed higher on the tests ON AVERAGE than condition 0, or do you mean that condition 1 gained more across the two time periods than condition 0, or do you mean something else?

Re: GLM

ballardw — Fri, 11 Sep 2020 14:52:56 GMT

If you have identification information to compare individual students scores this looks like a Paired T-test of means to determine the mean change in scores might well be appropriate.

Re: GLM

SteveDenham — Fri, 11 Sep 2020 15:13:28 GMT

Between @ballardw and @PaigeMiller , you should get the idea that a repeated measures analysis in GLM using the MANOVA approach is like swatting a fly with an AK-47. For this design, a paired t-test (using PROC TTEST) will more clearly answer your question.

As an aside, you probably shouldn't be using GLM for analysis of any repeated measures design in any case. The assumptions that go into the analysis are seldom met, the standard errors of the main effects in the model are miscalculated, and you need to have a pretty strong background in multivariate analysis to interpret the results. PROC TTEST for simple pre vs post tests in two groups, or PROC MIXED for more complex repeated measures are preferred.

SteveDenham

Re: GLM

zahidhasandipu — Sat, 12 Sep 2020 20:53:22 GMT

I am trying to interpret the results that are given. What could be the code to get a similar kind of results from any data set?

Re: GLM

PaigeMiller — Sat, 12 Sep 2020 21:45:57 GMT

@zahidhasandipu wrote:

I am trying to interpret the results that are given. What could be the code to get a similar kind of results from any data set?

This far too brief and far too vague a question for us to give an answer. Please elaborate.