Repeated Measures ANOVA test without Greenhouse Geisser and Huynh-Feldt P-values

[Ellie Wizard]

Good Afternoon,
I am hoping someone is able to help. I'll try and be as concise as possible to make it easier to understand.

I am a MSc cardiac sonographer looking at EF for the LV and FAC for the RV against global longitudinal strain (GLS) values for the RV and LV. The LV GLS and EF values are skewed and I've performed another test analysis on them.

I am using the STATA 16.1 software on Mac OS Catalina.

The values I've used are listed below.

* Example generated by -dataex-. To install: ssc install dataex
clear
input float gls_rv byte patient_id float rvfac byte chemostage
-14.5  1 43.7  0
  -14  1 54.7  1
-19.6  1 52.8  2
-18.5  2 59.2  0
-22.9  2 61.6  1
-29.8  3 47.9  0
-25.1  3 49.6  1
-21.8  3 41.4  2
-24.2  3   44  3
-19.1  4 46.9  0
-15.3  4 49.6  1
-19.3  4 48.5  2
-17.5  4 48.5  4
-19.5  4 48.6  5
  -21  4 50.6  6
-23.3  4 56.9  7
-18.5  4 51.7  8
-24.4  5 48.8  0
-20.4  5 42.9  1
-23.3  5 44.1  2
-19.2  5 40.6  3
-16.8  5 39.4  4
-16.1  5 54.5  5
-17.6  5 35.9  6
-21.8  5 41.5  7
-22.9  6 53.1  0
-22.3  6 58.1  1
-24.4  7 52.3  0
  -23  7 47.6  1
-24.8  7 52.8  2
-25.3  7 49.6  3
-21.1  7 52.2  4
-30.5  7 53.3  5
-26.7  7 48.7  6
-29.1  7 52.8  7
-20.5  7 57.7  8
-21.1  8 60.1  0
  -19  8 60.4  1
-13.1  9   51  0
-15.9  9 48.2  1
-11.3  9 50.5  2
-13.3  9 58.9  3
-14.5 10   53  0
-22.7 10   53  1
-23.8 10 63.7  2
-21.6 10 63.9  3
-22.7 11 60.2  0
  -19 11   51  1
-18.7 11 47.4  2
-21.3 11 55.1  3
-21.2 11 52.5  4
-20.9 13   59  0
-26.9 13 64.8  1
-23.1 13   45  2
-25.7 13 43.8  3
-25.6 13 57.6  6
-23.7 13 53.8  7
-23.3 14 43.6  0
-23.3 14 45.6  1
-19.6 14 44.3  2
-17.2 14 37.1  3
-22.4 14 48.4  4
-24.1 14 51.2  5
-17.1 14 49.1  6
-14.4 14 52.3  7
-22.4 15 57.1  0
-19.6 15 60.4  1
 -9.8 15 55.7  2
-22.4 15 57.3  3
-19.3 15   55  4
-20.9 16 60.6  0
-23.8 16 58.7  1
-23.5 16 55.7  2
  -21 16 50.4  3
  -28 16 46.6  5
-17.1 16 51.2  6
  -20 17 60.6  0
-19.1 17 54.5  1
  -19 17 46.1  3
-20.9 17 62.5  4
-14.6 17 54.7  5
-19.3 17 59.3  6
-21.6 18 48.6  0
-20.3 18 48.8  1
-20.9 19 57.7  0
-19.7 19 63.6  2
-23.6 19 58.5  3
-20.4 19 67.3  4
-23.1 19 54.6  5
-21.5 19 59.8  6
-25.3 19 58.4  7
-21.3 19 67.2  8
-23.3 19 67.5  9
-16.8 19 61.2 10
-29.1 20 50.9  0
-26.4 20 59.6  1
-25.8 20   58  2
-20.5 20 50.1  3
-20.5 20 52.7  4
-18.1 20 56.6  6
end

RV GLS and RAC are both normally distributed. Before running the ANOVA, I tested the assumption on sphericity using with rvfac gls_rv and chemostage:

mauchly rvfac gls_rv chemostage

I also ran this with patient_id. I was unsure as to whether it needed another variable.....

mauchly rvfac gls_rv patient_id chemostage

If i have done this correctly. it says that I cannot use the p-value (prob > F) and have to refer to the Greenhouse Guisser or Huynh-Feldt epsilon value instead.

For ANOVA I used:

anova gls_rv patient_id chemostage, repeated(chemostage)

I was originally told not to use the patient_id variable. However, it could not determine between-subject without identification. Furthermore after adding patient ID as presented above the greenhouse geisser or Huynh-Feldt values are not being displayed. I have attached png files above as I am unable to dataex mauchlys sphericity and ANOVA.

Would anyone know how to get those values?

Thank you
Ellie

[Joseph Coveney]

I think that your problem is that your repeated measurements grow too sparse as the interval lengthens. You're down to single observations at the longer intervals. See below.

.ÿ
.ÿversionÿ16.1

.ÿ
.ÿclearÿ*

.ÿ
.ÿquietlyÿinputÿfloatÿgls_rvÿbyteÿpatient_idÿfloatÿrvfacÿbyteÿchemostage

.ÿ
.ÿisidÿpatient_idÿchemostage

.ÿtabulateÿchemostage,ÿmissing

ÿchemostageÿ|ÿÿÿÿÿÿFreq.ÿÿÿÿÿPercentÿÿÿÿÿÿÿÿCum.
------------+-----------------------------------
ÿÿÿÿÿÿÿÿÿÿ0ÿ|ÿÿÿÿÿÿÿÿÿ19ÿÿÿÿÿÿÿ19.00ÿÿÿÿÿÿÿ19.00
ÿÿÿÿÿÿÿÿÿÿ1ÿ|ÿÿÿÿÿÿÿÿÿ18ÿÿÿÿÿÿÿ18.00ÿÿÿÿÿÿÿ37.00
ÿÿÿÿÿÿÿÿÿÿ2ÿ|ÿÿÿÿÿÿÿÿÿ14ÿÿÿÿÿÿÿ14.00ÿÿÿÿÿÿÿ51.00
ÿÿÿÿÿÿÿÿÿÿ3ÿ|ÿÿÿÿÿÿÿÿÿ13ÿÿÿÿÿÿÿ13.00ÿÿÿÿÿÿÿ64.00
ÿÿÿÿÿÿÿÿÿÿ4ÿ|ÿÿÿÿÿÿÿÿÿÿ9ÿÿÿÿÿÿÿÿ9.00ÿÿÿÿÿÿÿ73.00
ÿÿÿÿÿÿÿÿÿÿ5ÿ|ÿÿÿÿÿÿÿÿÿÿ7ÿÿÿÿÿÿÿÿ7.00ÿÿÿÿÿÿÿ80.00
ÿÿÿÿÿÿÿÿÿÿ6ÿ|ÿÿÿÿÿÿÿÿÿÿ9ÿÿÿÿÿÿÿÿ9.00ÿÿÿÿÿÿÿ89.00
ÿÿÿÿÿÿÿÿÿÿ7ÿ|ÿÿÿÿÿÿÿÿÿÿ6ÿÿÿÿÿÿÿÿ6.00ÿÿÿÿÿÿÿ95.00
ÿÿÿÿÿÿÿÿÿÿ8ÿ|ÿÿÿÿÿÿÿÿÿÿ3ÿÿÿÿÿÿÿÿ3.00ÿÿÿÿÿÿÿ98.00
ÿÿÿÿÿÿÿÿÿÿ9ÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿÿÿÿÿ1.00ÿÿÿÿÿÿÿ99.00
ÿÿÿÿÿÿÿÿÿ10ÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿÿÿÿÿ1.00ÿÿÿÿÿÿ100.00
------------+-----------------------------------
ÿÿÿÿÿÿTotalÿ|ÿÿÿÿÿÿÿÿ100ÿÿÿÿÿÿ100.00

.ÿ
.ÿexit

endÿofÿdo-file


.

Your -dataex- thingy captures only 100 of your 177 total observations and so there are only 10 levels of the chemostage variable present in the excerpt. From the screenshots in your post, there are nine more levels, and I doubt things get any better. I don't think that there's anything that you can do about this other than to restrict the number of repeated measurements subjected to analysis.