 积分
 11307
好友
记录
日志
相册
帖子
主题
分享
精华
威望 旺
钢镚 分
推荐 人

注册后推荐绑定QQ，之后方才可以使用下方的“用QQ帐号登录”。
您需要 登录 才可以下载或查看，没有帐号？立即注册
x
Through the use of a method novel to suicide research, an established procedure was replicated for scoring gendertyped attitudes and behaviors in which a single latent probability variable of identifying as male was generated from 16 genderdiscriminating variables (including not crying, physically fit, not moody, not emotional, liking yourself, fighting, and risk taking).6 Participants scoring 73% probability or higher of identifying as male (>1 SD above the mean) were coded as HTM.
Association of High Traditional Masculinity and Risk of Suicide Death Secondary Analysis of the Add Health Study
JAMA Psychiatry. Published online February 12, 2020. doi:10.1001/jamapsychiatry.2019.4702
参考文献6
Environmental and Genetic Influences on SexTyped Behaviors and Attitudes of Male and Female Adolescents
First Published December 1, 2001 Research Article
https://doi.org/10.1177/01461672012712003
SexTypicality Scale Probability of Being a Boy (PRBOY) for Boys and Girls
Variable Name  Content of Items  Direction of Response Pattern  CRYALOT  Frequency of crying  1 to 4; 4 = every day  MOODY  Frequency of moodiness  1 to 4; 4 = every day  NO_APPET  Frequency of poor appetite  1 to 4; 4 = every day  HONEST  Honestly answered questions  1 to 4; 4 = completely honest  TRBATTEN  Trouble paying attention  1 to 4; 4 = every day  BOTHER  Bothered by things  1 to 4; 4 = every day  PHYSICFT  How physically fit  1 to 5; 5 = strongly disagree  BADFIGHT  Past 12 months, serious fighting  0 to 3; 3 = five or more times  AEROBICS  Frequency of exercising  0 to 3; 3 = five or more times  SKATE  Frequency of rollerblading/cycling  0 to 3; 3 = five or more times  EMOTION  How emotional you are  1 to 5; 5 = strongly disagree  LIKESELF  Do you like yourself as you are  1 to 5; 5 = strongly disagree  LIVNOTHK  Live without thought for future  1 to 5; 5 = strongly disagree  SENSITIV  How sensitive to others’ feelings  1 to 5; 5 = strongly disagree  TAKERISK  Do you like to take risks  1 to 5; 5 = strongly disagree  UPSETPRB  Upset by difficult problems  1 to 5; 5 = strongly disagree 
We developed a measure of sex typicality of behaviors and attitudes for adolescents in Wave II of the Add Health study usingan approach similar in nature to work described by Lippa and Connelly (1990). The main objective was to construct a sextyped behaviors and attitudes score from the probability that an adolescent is male (or female) on the basis of participants’ responses to a set of questions.
It is similar to the approach used by Lippa and colleagues in that it retrofits a gender score to preexistingdata based on differential responses of males
and females but it differs in application. Instead of using probabilities of beingmale or female derived from larger psychological inventories, such as occupational preference inventories or the CPI (e.g., Lippa & Connelly, 1990; Lippa & Hershberger, 1999), our approach uses individual items and their contributions to the probabilities of being a boy.
This approach is less demanding of data and allows the construction of gender scales on a wider range of existing data sets. In applyingthis technique, we first selected a broad set of Wave II items from varyingbehavioral and attitudinal domains that showed sex differences in a response that could be attributed to individual preferences or behaviors. Care was taken not to select items where social restrictions presented males’ and females’ differential opportunities to participate (e.g., playing baseball).
Preliminary analysis usingthe core sample of Wave II identified 21 questions that were useful in discriminating boys from girls. Stepwise logistic regression was used to select a subset of these questions that significantly contributed to predictingthe logodds of beinga boy. Table 1 provides the variable names, content of items, minimum and maximum scores, and difference scores (differences in mean response for boys and girls expressed as a fraction of the standard deviation of the boy). When examiningthe signs precedingdifference scores in Table 1, bear in mind that the direction of these scores is affected by the direction of the responses for each item.
For example, response patterns of items PHYSICFT, EMOTION, LIKESELF, LIVNOTHK, SENSITIV, TAKERISK, and UPSETPRB ranged from 1 (strongly agree) to 5 (strongly disagree). Accordingly, the difference score for EMOTION (.44) does not indicate that females are less likely than males to agree that they are emotional. In contrast, they are more likely to do so. The unsigned average of Table 1’s difference scores was .37, indicating that responses to the individual items, although distinguishing between the sexes, were not dramatically different for males and females. The regression equation including these variables is as follows:
loge[pi/(1 – pi)] = 2.2098 + CRYALOT*(–1.2525)+ MOODY*(–0.2255)
+ NO_APPET*(–0.2022)
+ HONEST*(–0.2435)
+ TRBATTEN*(0.3109)
+ BOTHER*(–0.1194)
+ PHYSICFT*(–0.3676)
+ BADFIGHT*(0.6654)
+ AEROBICS*(–0.1331)
+ SKATE*(0.3056)
+EMOTION*(0.1217)
+ LIKESELF*(–0.2042)
+ LIVNOTHK*(–0.2171)
+ SENSITIV*(0.3175)
+ TAKERISK*(–0.1762)
+ UPSETPRB*(0.2025)
where pi is the probability of beinga boy for the ith adolescent. By solvingfor pi , individual probabilities were computed for each adolescent and used as the measure of sex typicality. The resultingvalues range from 0 to 1, with 0 denotingthe most femalelike and 1 the most malelike score. The coefficients in the above formula provide the specific contribution of each variable to the logodds of beinga boy while holdingthe contributions of other
variables in the equation constant. One implication of this is that intrascale collinearity, which increases internal measures of reliability such as Cronbach’s alpha, is limited by design. Measures of internal reliability, such as Cronbach’s alpha, are appropriate measures of reliability for scales usingmultiple items to assess a single domain. The approach used to construct PRBOY draws on measures from multiple domains that each contribute to the classification of individuals as males and females and is linked to our definition of gender not being limited to one or a few domains.
Increasingthe number of domains assessed by the scale increases PRBOY’s ability to correctly classify adolescents by sex and provides explicit recognition that sex differences exist across domains of behavior and attitudes. Accordingly, a more relevant issue is whether the full model functions as intended. To determine if the PRBOY correctly classifies adolescents by sex, we used receiver operatingcharacteristic (ROC) analysis. ROC measures of accuracy have been used in medical testing, information retrieval, weather prediction, (Swets, 1988), psychology (Swets, 1973), and epidemiology (Erdreich & Lee, 1981). To construct an ROC curve, we computed the probability of a true positive (probability that adolescent is a boy when truly a boy) versus the probability of a false positive (probability that adolescent is a boy when truly a girl) using various cutpoints between 0 and 1. The area under this curve measures the probability of a correct ranking. It has been shown that this area measure is the same quantity that is
estimated by the nonparametric Wilcoxon statistic (Hanley & McNeil, 1982). A value of 0.5 would indicate correct classification is the same as chance. Values of area under ROC curves range from 0.90 to 0.98 for diagnosis from applications of CT and chest xray films, 0.80 to 0.90 for mammography, and 0.75 to 0.90 for weather prediction (Swets, 1988). Calculated on the core sample (N = 8,421), which was used to derive the formula for the
PRBOY, analysis showed our model correctly discriminates between a randomly chosen boy and girl 81.7% of the time. To determine if this ROC was inflated due to beingcalculated on the same sample used to derive the PRBOY, the ROC was recalculated on the noncore sample respondents with complete PRBOY data (N = 4,789).

