Native American and Multi-Race Males: Binary Logistic Regression Analysis Summary
Attempted Suicide (Multiple Times and/or Received Medical Care) In Past Year
Independent

Variables

/

Enter Regression Results: Odds Ratios (Relogit Values1)

Native American And All Multi-Race Males

/

Count, %F, Correlation, Odds Ratio

Not Weighted / Weighted2

Statistical Significance

a = <= 0.000, b = 0.001-0.009, c = 0.010-0.049
d = 0.050-0.099, e >= 0.100
2X2 listing order: Three groups together, and Native (One Race), Native (Multi-Race), Multi-Race (No Native),, respectively.

Not Weighted

D. Variable Frequency
17 / 595 – 2.9% /

Weighted

Dependent Variable Frequency
25 / 579 = 4.3%
NW
1:10 / NW
AICr / NW

AICr

/ W
1:10 /

W

AICr / W
AICr / W-1
1:10 / W-1

AICr

/ W-1

AICr

/

2X2: NW

/ 2X2: W
Gay Harassment,
School, Year
/ 10.4a / 10.9a / 11.4a / 5.5b / 7.6a /
8, 14.0%, 0.218: 9.6a
3, 16.7%, 0.235: 11.6a
3, 15.0%, 0.300: 22.6a
2, 10.5%, 0.141: 5.0d /
11, 18.3%, 0.242: 8.7a
2, 11.1%, 0.122: 4.4d
5, 23.8%, 0.361: 17.5a
5, 22.7%, 0.289: 10.9a
Slapped/Hit at
Home by Adult, Year
/ 6 counts shared with
gay harassment.3 / 9 counts shared with
gay harassment. 3 / 9 counts shared with
gay harassment. 3 / 8, 9.1%, 0.156: 5.5a
4, 11.8%, 0.216: 9.6b
2, 9.1%, 0.165: 6.2c
2, 6.3%, 0.080: 2.6e / 13, 13.4%, 0.200: 6.1a
5, 15.2%, 0.261: 11.7a
3, 13.6%, 0.168: 4.3d
5, 11.9%, 0.169: 4.5c
Sniff Glue, etc.,
Life, times: 10+
/ 24.8a / 20.2a / 25.1a / 24.8a / 15.9a / 12.2a / 4, 18.2%, 0.180: 9.6a
0, 0.0%, -0.030e
1, 20.0%, 0.200: 11.8c
3, 33.3%, 0.385: 30.0a / 9, 31.0%, 0.302: 15.0a
0, 0.0%, -0.038e
2, 33.3%, 0.302: 15.5a
7, 50.0%, 0.568: 65.0a
Steroid Use,
Life, times: 1+
/ 10.7b / 8.2b / 3, 13.0%, 0.123: 5.6b
1, 8.3%, 0.076: 3.6e
1, 16.7%, 0.177: 9.3c
1, 20.0%, 0.159: 9.1c / 3, 14.3%, 0.095: 4.1c
1, 9.1%, 0.061: 2.7e
2, 40.0%, 0.308: 16.5a
1, 20.0%, 0.104: 5.0e
Fights, Year, #: 6+,
<=1 at School
/ 10.0b / 23.4a / 21.6a / 22.9a / 23.4a / 26.4a / 28.1a / 3, 14.3%, 0.131: 6.7b
1, 14.3%, 0.119: 6.7d
1, 16.7%, 0.177: 9.3c
1, 12.5%, 0.112: 5.1e / 6, 25.0%, 0.212: 9.4a
2, 22.2%, 0.207: 10.4b
2, 40.0%, 0.338: 20.7a
2, 20.0%, 0.160: 6.1c
Medical
Check-Up, Met4 / 0.127b / 2, 0.8%, -0.104: 0.180c
1, 1.0%, -0.090: .337e
0, 0.0%, -0.152d
1, 1.3%, -0.086: .289e / 3, 1.2%, -0.130: 0.176b
1, 1.0%, -0.125: .165e
0, 0.0%, -0.195c
2, 2.3%, -0.095: .344e
Drugs-Alcohol
HCN: Met / 12.9b / 16.8b / 16.0b / 12.2b / 2, 14.3%, 0.106: 6.3b
0, 0.0%, -0.024e
0, 0.0%, -0.014e
2, 25.0%, 0.262: 15.0a / 3, 23.1%, 0.140: 7.4b
0, 0.0%, -0.025e
0, 0.0%
3, 37.5%, 0.306: 16.9a
P-E Problems HCN,
Met/Unmet / 5.2c / 7.5b / R / R / R / 5, 16.1%, 0.187: 8.8a
0, 0.0%, -0.026e
2, 22.2%, 0.306: 19.7a
3, 18.8%, 0.271: 13.3a / 7, 20.0%, 0.196: 7.3a
0. 0.0%, 0.025e
4, 30.8%, 0.377: 17.6a
3, 16.7%, 0.171: 5.3c
Constant / 0.016a / 0.012a / 0.010a / 0.018a / 0.017a / 0.024a / 0.018a / 0.014a / 0.014a

Hosmer &Lem. Test5

/ -- / -- / 0.333 / -- / 0.620 / 0.886 / -- / 0.624 / 0.208

-2Log Likelihood5

/ 141.7 / 133.3 / 124.6 / 170.4 / 157.7 / 143.2 / 170.4 / 158.9 / 147.1

Notes and Abbreviations:

NW, W: Not Weighted, Weighted.

NW-1, W-1 : Same regression as preceding one, with one variable removed indicated by “R”

D. Variable Frequency : Frequency for the dependent variable: “Attempted Suicide (Multiple Times and/or Received Medical Care) In Past Year”.

1:10 : Following the rule “1 predictor per 10 events” for logistic regression.

AICr : Including/excluding variable is based on AIC (Akaike Information Criterion) value differences being statistically significant. BIC (Bayesian Information Criterion) value differences are generally significant for the second least significant variable in regression models with 4 predictors.

P : A significant variable that could have substituted for one of the less significant variables used depending on variable selection/removal criteria used.

1.  Relogit by King et al. (1999, 2001: http://gking.harvard.edu/stats.shtml) is available as a plug-in for the Stata 8 Statistical Software. It produces corrected coefficients (converted into Odds Ratios). Logit produces correct coefficients only if the “1 predictor per 20 events” rule is followed.

2.  2X2 values generated with the maximum removal of study subjects in a logistic regression with 50 variables. Actual counts in regression would sometimes be higher.

3.  The available research leads to the suggestion that males known to be gay, or suspected to be gay on the basis of gender nonconformity (the most common situation) are more likely to be abused and also physically assaulted in their homes, at school, or on their way to school.

4.  For this group of males, 334 out of 579 males (57.7% of males) reported not having had a yearly medical check-up or a sports physical examination and 22 out of 24 attempters (weighted) are in this “more serious” “attempted suicide” category. Not weighted: 15 out of 17 suicide attempters are in the 347 / 595 males (58.3% of males) who did not report having a medical check-up. Are males in the “at risk category more likely to be in the lower SES? Not weighted, 10 out of the 18 suicide attempters (55.5%) are in the lower-middle to lower SES. Weighted, 10 out of 26 attempters (38.5%) are in this category.

5.  Hosmer & Lem. Test: Hosmer and Lemeshow Test. The statistical non-significance – as given - suggests that the logistic regression model is good. Values generated on the basis of Logit regression values (SPSS 11). They are not available with the Relogit plug-in program in Stata 8.