Study design, setting and participants
The methodology of this cross-sectional study, which was conducted from June to November 2013, has already been described in Amarasuriya et al. [9, 29]. The study population consisted of undergraduates enrolled in a Sri Lankan university.
The study sites included five of the six undergraduate faculties of the University of Colombo, namely the Faculties of Arts and Education, Law, Management and Finance, Medicine, and Science, as well as the School of Computing, which is an affiliated institute of the university. The sampling strategy, which involved clustered sampling, aimed to produce as large a sample as possible by approaching all those attending a lecture (the cluster) which was common for each year of study in the research sites. In the case of the Faculty of Arts in which students had varied subject combinations, lectures with the largest student cohorts were approached for data collection. Data was not collected in the Faculty of Education, as the second and third year students of this faculty attended lectures in the Faculty of Arts, and as only fourth year students had lectures exclusively at this faculty. The strategy of systematically approaching students from all faculties and years of study during the identified lectures was considered to reduce any bias in sampling.
Depression and depression literacy of the undergraduates was examined using a paper-based dual language questionnaire that was available in two versions, i.e., as English-Sinhala and English-Tamil versions. Participants could choose their preferred version. A vignette which described an undergraduate named ‘Z’, who exhibited symptoms of Major Depression as per the Diagnostic and Statistical Manual of Mental Disorders-IV, was provided as the problem trigger. The present paper examines the predictors of the undergraduates’ personal help-seeking intentions. Therefore, it only describes the questionnaire items relevant to this aim. The complete depression literacy questionnaire, including the depression vignette that was used and details regarding its development, have been previously published [9, 29]. The questionnaire has been provided as Additional file 1 for the convenience of the reader.
The undergraduates’ help-seeking intentions were elicited in relation to the depression vignette using the following open-ended question: “If you have this problem, what would you do?” Multiple responses were possible.
This was examined using the following open-ended question: “What do you think is wrong with ‘Z’?” Multiple responses were possible. Amarasuriya et al.  categorised these responses as follows: recognised as depression, recognised using other mental health-related labels (pertaining to the label categories “mental illness”, “stress/pressure/mental suffering”, “mental issue”), and not recognized (where responses were not relevant to either of the two previous response categories).
Screening positive for major depression
The Patient Health Questionnaire-9 (PHQ-9) is a self-administered measure which consists of nine items, based on the nine symptoms of Major Depression in the DSM-IV (Kroenke & Spitzer, 2002). This measure instructed respondents to rate the degree to which they were bothered by the symptoms presented in the measure during the previous two weeks, using a four point scale consisting of the following rating options: Not at all, Several days, More than half the days, Nearly every day. These rating options were scored from 0 to 3 respectively. Therefore, a maximum score of 27 could be obtained. If five of the nine symptoms were found to be present for More than half the days or Nearly every day, then a diagnosis of Major Depression was given. However, for these diagnoses to be given, either the symptom relating to depressed mood or anhedonia had to be present. If the symptom relating to suicidal thoughts was present at all, it was then considered within the symptom count for these diagnoses.
The questionnaire examined the undergraduates’ stigmatising attitudes towards those with depression, by examining their negative attitudes towards ‘Z’ using the Personal Stigma Scale, and their willingness to interact with ‘Z’ using the Social Distance Scale.
The Personal Stigma Scale is a component of the Depression Stigma Scale . This study used a version of this scale adapted for use among young persons between 12 and 25 years of age (Jorm & Wright, 2008). The scale consisted of seven statements and asked participants to indicate how much they agreed or disagreed with the statements, by using a five-point rating scale (scored from 1 to 5) consisting of the following rating options: Strongly agree, Agree, Neither agree nor disagree, Disagree, Strongly disagree. The items were reverse-scored so that higher scores indicated a higher level of personal stigma. A total score of 35 could be obtained.
The Social Distance Scale which was used is based on a social distance scale for adults , that was later adapted for use among a population of young persons between 12 and 25 years of age (Jorm & Wright, 2008). The scale consists of five items which ask participants to indicate their willingness to do the activities which are listed with ‘Z’, using a four-point rating scale (scored from 1 to 4), consisting of the following rating options: Yes, definitely, Yes, probably, Probably not, Definitely not. Higher scores on the scale indicate lesser willingness to interact with ‘Z’ and hence, greater desire for social distance from ‘Z’. A total score of 20 could be obtained.
Amarasuriya et al.  found that the Personal Stigma scale consisted of two dimensions of stigma, i.e., the Weak-not-Sick and Dangerous-Undesirable dimensions (consisting of three items each), while the Social Distance scale consisted of one dimension (consisting of five items), which was labelled as Social Distance. This led to the construction of three subscales which reflected these three dimensions of stigma (see Amarasuriya et al.  for details regarding construction of the scales, factor loadings of scale items, reliabilities of scales and their limitations relating to low reliability estimates).
Exposure to problem and help-seeking
Three closed questions were used to examine the undergraduates’ previous experiences of the problem and their related help-seeking practices. Participants were asked if anyone in their family or close circle of friends had experienced a problem like ‘Z’s (response options: Yes, No, Don’t know), if they had ever had a problem like ‘Z’s (response options: Yes, No, Don’t know), and in the case of responding as ‘Yes’ or ‘don’t know’ to the latter, if they had dealt with the problem on their own, without getting help from others (response options: Yes, Tried first but got help later, No).
The following variables were examined: faculty of study (open-ended question), gender, year of study, age, religion (response options: Buddhist, Hindu, Islamic, Christian, Roman Catholic, Atheist, Other) and residence (response options: at home, hostel, rented place, home of friend/relative, other).
The questionnaires were distributed to the undergraduates during lectures. Participation was voluntary and anonymous. Implied consent procedures were used, where consent to participate was implied when a filled questionnaire was returned. The undergraduates took approximately 20 min to complete the questionnaire.
The aggregates for all types of professional and informal help that the undergraduates identified as those they would seek if personally affected by the problem were calculated. The category “person not specified” was excluded from the aggregate for informal help due to the ambiguity of this response. The aggregate for help-seeking intentions relating to self-help strategies was not calculated, as there were a range of differing responses, some of which may have been unhelpful when dealing with depression. The correlation between the aggregates for intentions to seek professional and informal help was also found using the phi correlation coefficient.
Binary logistic regression models were used to examine the predictors (IVs) of each of the aforementioned help-seeking intentions of the undergraduates (DVs) (i.e., professional help, informal help and self-help strategies). The following categorical variables were simultaneously entered into binary logistic regression models to examine the predictors of each of these types of help-providers/strategies of help-seeking, where the variable sub-categories which are italicised were the reference categories for each of the respective variables: gender (male, female); faculty of study (Medicine, Arts and Education, Law, Management and Finance, Science, School of Computing); year of study (1st year, 2nd year, 3rd year, 4th year, 5th year Medicine); age category (18–20 years, 21–23 years, 24 years and above); ability to recognise the problem in vignette (not recognised, recognised as depression, recognised using other mental health-related labels (pertaining to the label categories “mental illness”, “stress/pressure/mental suffering”, “mental issue”); if the respondents screened positive for Major Depression as per the PHQ-9 (no, yes); if the problem in the vignette had been experienced by family or friends (no, yes, don’t know); if the problem was personally experienced (no, yes, don’t know); and from those who responded as yes or don’t know to the latter question, if they had sought help for the problem (help not sought, tried first but got help later, help sought; those who did not indicate personal experience of the problem were included in the analysis, but were dummy coded as a not relevant category). The stigma scale scores, which were continuous variables, were entered into the model in relation to the Weak-not-Sick, Dangerous-Undesirable and Social Distance scales that have been described in Amarasuriya et al. . As simultaneous regression analysis models were used, this allowed for an examination of each of these variables as predictors of the undergraduates’ help-seeking intentions while adjusting for all the other variables entered into the model. The analyses were also adjusted for the participants’ religion, residence and language of response. Given the large number of predictors entered into each of the regression models, the p < .01 significance level was used to reduce the Type I error rate.
In order to examine the effect sizes of these associations, in instances in which the ORs were < 1, these were converted to 1/OR to place all effects in a common frame of reference . ORs of 1.5 were considered as small effect sizes, 2.5 as medium effect sizes, 4 as a large effect sizes and 10 as very large effect sizes . The results were interpreted in relation to these effect size estimates. As described in Mackinnon , the dichotomization of continuous scaled predictors for use in logistic regression analyses reduces the power of these scaled predictors. Therefore, as recommended by Mackinnon, in the case of the stigma scores which were continuous scaled predictors, Interquartile Odds Ratios (IQOR) which retain the full power of these predictors were calculated. This placed these variables within a common frame of reference with the other predictor variables when examining their effect sizes.
It must be noted that problem recognition as a predictor of intentions to seek help from professional help-providers has already been examined in Amarasuriya et al. . The present analysis adjusted for a larger number of variables (stigma, exposure to problem and whether help was sought if problem was personally experienced), and also examined whether problem-recognition is a predictor of the undergraduates’ intentions to seek help from informal help-providers or to engage in the self-help strategies identified.
Data was analysed using the Statistical Package for the Social Sciences, version 23.