topic Re: Proc mixed help in Statistical Procedures

Proc mixed help

HBa — Mon, 27 Jan 2020 17:23:44 GMT

I am learning to proc mixed. So my following question is hypothetical and any help is appreciated:

Supposed that I am comparing three age different un-equal groups of runners who ran with different shoe types and would like to see if they are significantly different in terms of their speed.

Then, my data can have the following variables: subj_id, group name, shoe type, speed of running.

proc mixed data=temp;

class subj_id group_name shoe_type;

model speed_of_running= group_name shoe_type group_name*shoe_type;

random subj_id;

run;

I have the following questions:

Is my syntax above correct for this application? Do I need to randomize subj_id?
Can I alter proc mixed so that I can see if each group is also different? For instance, if I have three groups of A, B, and C, can I make changes to see if A and B are different, B and C are different, A and C are different.
Can I alter proc mixed to see which group ran faster on average?

Thanks

Re: Proc mixed help

PaigeMiller — Mon, 27 Jan 2020 17:48:38 GMT

You didn't really explain the design of your study, and I would need to know that to determine if your code is correct. So please explain the design of your study, so we can understand how subjects get assigned to groups and shoe_types and how often they run once assigned.

Specifically, does each subject_id run and is timed a single time? Or does each subject_id run and is timed more than once? Does each subject_id wear different shoe types throughout the study? I assume each subject can only appear in one group.

With regards to your last two bullet items, the LSMEANS statement will compute the means for you and do statistical testing, so that you can compare groups and compare shoe types.

Re: Proc mixed help

HBa — Mon, 27 Jan 2020 17:53:01 GMT

Thank you PaigeMiller

Let's consider that i have 35 subjects in the three unequal groups. the subject_id variable goes from 1 to 35. All subjects run 5 minutes with each of the shoe types while i measure their speed. For instance, it can be running bare feet, sport shoe and dress shoe. the order of shoe type for each subject is randomly assigned. so, technically, each subject can only appear in one group but is tested three times for the three conditions of shoe type.

PS:can you also tell me how the code would have been different if

a- each test was repeated three times. i.e. each subject belongs to one of the three age groups but runs 9 times (three times with each shoe).

b- can you also tell my how the code would have been different if the subject appeared in more than one group. For instance if my test groups instead of being different in age, they were the same subject by were were tested before and after having breakfast, or were tested morning, afternoon and night)

Re: Proc mixed help

PaigeMiller — Mon, 27 Jan 2020 18:06:02 GMT

It seems to me then that you want this RANDOM statement

random subj_id(group_name);

since subject_id is nested within group_name.

Re: Proc mixed help

HBa — Mon, 27 Jan 2020 18:17:37 GMT

So when i use lsmeans, waht does the p value referes to?

So if i have gorups A, B and C and i include a statement lsmean(group_name), i get something like the following:

Effect Group_name Estimate Error DF t Value Pr > |t|

Group_name A .......

Is the p value refering to the difference in mean of A as compared to all other groups?

thanks

Re: Proc mixed help

PaigeMiller — Mon, 27 Jan 2020 18:22:04 GMT

Please show us the actual code used, and the actual table produced by that code.

Re: Proc mixed help

HBa — Mon, 27 Jan 2020 18:55:53 GMT

proc mixed data=temp;

class subj GROUP ShoeType ;

model SPEED= GROUP ShoeType GROUP*ShoeType ;

random subj (GROUP);

lsmeans GROUP;

run;

$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$

Kdisart YA 10:25 Monday, January 27, 2020 22

The Mixed Procedure

Model Information

Data Set WORK.TEMP

Dependent Variable SPEED

Covariance Structure Variance Components

Estimation Method REML

Residual Variance Method Profile

Fixed Effects SE Method Model-Based

Degrees of Freedom Method Containment

Class Level Information

Class Levels Values

subj 26 3 4 5 7 8 9 10 11 12 13 15 16

17 19 20 21 22 23 24 25 26 27

29 30 31 32

GROUP 3 AM OA YA

ShoeType 3 1 2 3

Dimensions

Covariance Parameters 2

Columns in X 16

Columns in Z 26

Subjects 1

Max Obs per Subject 234

Number of Observations

Number of Observations Read 234

Number of Observations Used 234

Number of Observations Not Used 0

Iteration History

Iteration Evaluations -2 Res Log Like Criterion

0 1 1001.35302689

1 1 954.65672525 0.00000000

Convergence criteria met.

Covariance Parameter

Estimates

Cov Parm Estimate

subj(GROUP) 1.5245

Residual 3.0052

Fit Statistics

-2 Res Log Likelihood 954.7

AIC (Smaller is Better) 958.7

AICC (Smaller is Better) 958.7

BIC (Smaller is Better) 961.2

Type 3 Tests of Fixed Effects

Num Den

Effect DF DF F Value Pr > F

GROUP 2 23 1.09 0.3540

ShoeType 2 202 13.11 <.0001

GROUP*ShoeType 4 202 2.78 0.0280

Least Squares Means

Standard

Effect GROUP Estimate Error DF t Value Pr > |t|

GROUP AM 5.5424 0.5565 23 9.96 <.0001

GROUP OA 5.4719 0.4311 23 12.69 <.0001

GROUP YA 4.6901 0.4311 23 10.88 <.0001

$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$

I also receive the following when I type lsmeans GROUP*ShoeType

$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$

Least Squares Means

Standard

Effect GROUP ShoeType Estimate Error DF t Value Pr > |t|

GROUP*ShoeType AM 1 4.7795 0.6489 202 7.37 <.0001

GROUP*ShoeType AM 2 5.9108 0.6489 202 9.11 <.0001

GROUP*ShoeType AM 3 5.9370 0.6489 202 9.15 <.0001

GROUP*ShoeType OA 1 5.3605 0.5026 202 10.67 <.0001

GROUP*ShoeType OA 2 4.8596 0.5026 202 9.67 <.0001

GROUP*ShoeType OA 3 6.1958 0.5026 202 12.33 <.0001

GROUP*ShoeType YA 1 3.6963 0.5026 202 7.35 <.0001

GROUP*ShoeType YA 2 4.3853 0.5026 202 8.72 <.0001

GROUP*ShoeType YA 3 5.9887 0.5026 202 11.92 <.0001

Re: Proc mixed help

PaigeMiller — Mon, 27 Jan 2020 20:21:37 GMT

Pr>|t| indicates if the statistical test for the hypothesis that estimate for that LSMEAN is equal to zero. The standard interpretation is that when this value is less than 0.05, then the estimate is not equal to zero. For group AM, the LSMEAN is 5.5424, which according to this statistical test, it is significantly different than 0.

To compare group A to group B, use

lsmeans group_name/lines;

I see in the documentation, there is no lines option in PROC MIXED, but there is one in PROC GLIMMIX, which for this design and analysis ought to give the same results as PROC MIXED.

Re: Proc mixed help

HBa — Wed, 29 Jan 2020 16:50:23 GMT

Thank you for your response. so, if variable "YOA" refers to the three groups that are called AM, OA and YA and variable "cond" refer to the condition (e.g. shoe type) and include conditions 1, 2 and 3. when i include lsmeans YOA*cond/lines; in my codes, i see the following:

T Grouping for YOA*cond Least Squares Means (Alpha=0.05)

LS-means with the same letter are not significantly different.

YOA	cond	Estimate

OA	3	15.9907		A
				A
OA	1	14.4178	B	A
			B	A
OA	2	13.7219	B	A
			B	A
AM	2	13.2346	B	A
			B	A
AM	3	12.613	B	A
			B	A
AM	1	11.5453	B	A	C
			B		C
YA	3	10.5709	B		C
					C
YA	1	7.4639			C
					C
YA	2	7.0647			C

does this mean that under condition 3 (i.e. the same shoe type), OA and AM were not different? also, AM conditions 1 and 2 were significantly different? In other words, which column do I have to look at?

thanks