Endocrine Disrupting Chemicals and Neurodegeneration

[mprec001]

Descriptive Statistics for categorical variables:

Gender
mean: 1.362869198
median: 1
mode: 1
SD: 0.481845186
variance: 0.232174784
Reference coding: NHANES data lists 1=male, 2=female

Education
mean: 3.181434599
median: 3
mode: 4
SD: 1.376556611
variance: 1.894908103
Reference coding: NHANES data lists 1=less than 9th grade, 2= 9-11, 3= High School/GED, 4=Some College/AA, 5= College Gradute

Presence of Phantosmia (Phantom Odor)
mean: 1.924050633
median: 2
mode: 2
SD: 0.265477754
variance: 0.070478438
Reference Coding: NHANES data lists 1=yes, 2=no

Smoking Status
mean: 2.278481013
median: 3
mode: 3
SD: 0.933419659
variance: 0.871272259
Reference Coding: NHANES data lists 1-Everyday, 2-Some days, 3-Not at all.

Graphs for all analysis 1variables to be uploaded next.

[mprec001]

Distribution graphs for the following variables:

Age - Continuous
Education - Categorical
Gender - Categorical
Phantom Odor - Categorical
Smoking Status - Categorical

The rest of the continuous variables are to follow. I will be running the graphs in R, as the function to create the stem and leaf plots in excel are not suited to larger ranges of numbers.

From my distributions so far, it does not appear I have many subjects that exhibit phantom odor (phantosmia), this may affect this specific analysis.

[mprec001]

Distribution graphs for the following variables:

Cadmium Urine Levels: Continuous
Triglycerides: Continuous
Cholesterol: Continuous
Glucose: Continuous
Systolic Blood Pressure: Continuous

The document contains the stem and leaf plots for each variable pre-discretization using R and the histogram post-discretization using Excel.

[mprec001]

Update:

I decided to use a different discretization scheme for my variables. Using a arbitrary binary cutoff for most of them may not lead ideal results, so I am using already established cutoffs to discretize variables as opposed to a binary 0,1 cutoff. Update of my variables and descriptive stats to follow.

[cwyoo]

Please justify your procedure. You may also look into dealing with values that are too high or too low (outliers).

Update:

I decided to use a different discretization scheme for my variables. Using a arbitrary binary cutoff for most of them may not lead ideal results, so I am using already established cutoffs to discretize variables as opposed to a binary 0,1 cutoff. Update of my variables and descriptive stats to follow.

[mprec001]

Attached are the 1 hr, 2 hr, and 4 hr run for all the health variables, heavy metal level variables, and brain health indicator variables. The excel sheet with discretization procedure, descriptive statistics, and stem-leaf and histogram to follow in order to justify my procedure.

Comments on graphs: The 3 runs appear to have the same nodes. The same d-separation and d-connectivity is evident between runs 2 and 4. There are differences between runs 2 and 4 in comparison to 1, but I attribute the differences to the run time used. I am waiting on a 6 hr run to finish and will post it as well when it is complete.

[mprec001]

6 hr run attached.

[mprec001]

In response to the questions I posted on the discussion board:

MARK: "In examining the different networks I have run at different times, I am finding the same d-connectivity, but a slight difference in the network score. Is this due to minor variations in the relationships with other variables? "

DR. YOO: "Are you saying the two different networks are equivelent? If so, they should have the same score. If the scores are different, then they cannot be equivelent, i.e., there should be different d-connected/d-separated relationships. Please post the two networks so we can go over with examples."

MARK: I re-examined the networks, and the 2 hr run has a different d-connected/d-separated relationship. The 1 hr, 4 hr, and 6 hr runs have all the same scores, which would make them equivalent. The 2 hr run, however, has a slightly different score, which is due to a difference the the d-connected/separated relationships.