Search or view whole report

View PDF

Method

Survey design

A copy of the survey instrument is presented in Appendix A.⁴ The survey was conducted by telephone interviews using a revised version of the survey developed for the Bega Valley pilot study (LJF 2003). In designing the pilot survey, the Law and Justice Foundation of NSW (LJF) drew on some of the recent legal needs surveys conducted in the United Kingdom (Genn 1999; Genn & Paterson 2001).

Measurement of legal events

As already noted, legal events were defined as events that have a potential legal resolution, whether or not the individual is aware of the legal consequences. As a result, participants were asked whether or not they had experienced specific events but were not required to determine whether or not these constituted legal needs or legal problems. Furthermore, the events were presented in context and in some detail to assist respondents to identify whether or not they had experienced these events, regardless of their level of legal knowledge. For example, rather than being asked whether they had a family law problem, participants were asked more specific questions, such as whether they had any problems with residence or contact arrangements for their children.

Thus, the term 'legal events' rather than the term 'legal needs' was adopted in the present report. The survey included as legal events:

1. events that are generally considered to be legal problems (e.g. instances of discrimination, criminal charges)
2. events that are generally not considered to be legal problems but involve clear legal consequences (e.g. buying or selling a house, making a will)
3. events that potentially have legal implications or remedies, but may not always be recognised as such (e.g. inadequate medical treatment, dispute with neighbour).

The survey examined legal events that occurred in the 12 months prior to the survey. Although other surveys have sometimes examined longer reference periods (e.g. Genn 1999; Genn & Paterson 2001; NCC 1995; Pleasence et al. 2004b), a shorter time period was adopted in the present study to maximise accurate recall of legal events.

Classification of legal events

While the survey concentrated on events that would usually be considered civil law issues, it also asked about a number of criminal law and family law issues. The classification of legal events adopted here was developed by the LJF and first reported by Scott et al. (2004).⁵ The present classification groups legalevents under three broad areas of law—civil, criminal and family law—and within more specific legal event groupings under each of these broad areas of law.

As presented in Table B1 in Appendix B, the survey identified 101 different legal events, which were classified as follows:

• 76 civil law events
• 16 criminal law events
• nine family law events.⁶

The civil law events were categorised into the following 11 legal event groups:

1. accident/injury (car accidents, work injury, personal injury)
2. business (problem as a landlord/business owner)
3. consumer (problem re superannuation, goods/services, financial institutions, insurance or lawyers)
4. credit/debt (problem re bill/debt, credit rating, money owed or being a guarantor; bankruptcy)
5. education (unfair exclusion, Higher Education Contribution Scheme (HECS) issue, bullying/harassment)
6. employment (problem re conditions, termination, harassment/mistreatment or discrimination)
7. government (problem re pension/benefit, government/disability/community services, taxation/debt, freedom of information, immigration, local council, non-traffic fines, immigration detention or legal system)
8. health (problem re psychiatric hospitalisation, medical treatment, disability facilities, non-government disability services or mental health care issue)
9. housing (buying/selling home; neighbour dispute; homelessness; problem re tenancy, home ownership, strata title, caravan/home estate, boarding house/hostel, retirement home/village or nursing home)
10. human rights (non-employment related discrimination on the basis of marital status, age, gender, religion, sexuality, ethnicity or disability)
11. wills/estates (making/altering a will, being the executor of a deceased estate, dispute over will/deceased estate, executing a power of attorney).

The criminal law events were categorised into the following three legal event groups:

1. domestic violence (victim or alleged perpetrator of family or household violence)
2. general crime (unfair treatment by police, criminal charge, problem re bail/remand, police failing to respond to a crime, victim of assault or stolen/vandalised property, problem in prison/juvenile detention)
3. traffic offences (loss of driver's licence, other traffic offence).

The family legal events were not grouped further. They included problems regarding residence and contact arrangements for children and grandchildren; problems regarding child support payments; child protection issues; issues involving fostering, adoption and guardianship of children, disabled people and elderly people; divorce/separation; and disputes regarding matrimonial property.

Survey description

The survey instrument was divided into four sections. The first section, entitled Screening and introduction (questions S1 to S6):

• identified the purpose of the survey
• identified the voluntary and confidential nature of the survey
• enabled the interviewer to check eligibility criteria (e.g. age, language spoken).

The second section, Your problems (questions 1 to 57):

• identified the legal events experienced in the 12 months prior to the survey⁷
• for participants who reported experiencing more than three legal events in the previous 12 months, identified the three most recent legal events
• included relevant filtering questions (questions 1, 3, 4, 6, 8, 11, 13, 15, 17, 18, 26, 28, 30, 33, 34, 36, 40, 41, 45, 52) to ensure that participants were only asked about legal events that they had the potential to experience in the previous 12 months (e.g. only participants who had been employed during the reference period were asked whether they had experienced employment-related legal events; only small business owners were asked whether they experienced any problems related to running a small business).⁸

The third section, What you did about your legal issues (questions 58 to 77), concentrated on the responses to, and the outcomes of, the three most recent legal events, identifying:

• whether the participant sought help
• sources of assistance
• barriers to accessing assistance
• satisfaction with assistance
• whether the matter was resolved
• satisfaction with the outcome.

The fourth section, Background information (questions 78 to 83), examined sociodemographic information including country of birth, Indigenous status, personal income, English language proficiency and language used to conduct the interview.

Sample

A two-stage sample design was used. The first stage involved the selection of disadvantaged areas to be sampled, and the second involved the sampling of individuals from those disadvantaged areas.

Selection of disadvantaged areas

The sample was drawn from six LGAs in NSW.⁹ Three of the six LGAs, Campbelltown, Fairfield and South Sydney, are suburban areas within Sydney. The other three LGAs, Newcastle, Nambucca and Walgett, are outside Sydney. Newcastle LGA is a large provincial centre within the Hunter Statistical Division (SD). Nambucca LGA is a rural area within the Mid-North Coast SD, surrounding the small town of Nambucca, and including a number of other small towns such as Macksville and Bowraville. Walgett LGA is a remote area within the North Western SD, which includes the small towns of Walgett, Lightning Ridge and Collarenebri.

The total sample size was 2431 with approximately 400 residents aged 15 years or over being surveyed from each LGA. Table 2.1 shows the sample and population numbers for each LGA. On average, the sample drawn from each LGA made up 0.5 per cent of the LGA population aged 15 years or over.

The selection of LGAs was based on the following considerations:

1. geographic diversity
2. socioeconomic disadvantage
3. cultural and linguistic diversity
4. population size.

Table 2.1: Sample and population size of each LGA, 2003

SD	LGA	Population no. (15+ years) a	Sample no.	Sample as % of population
Sydney	Campbelltown	113 459	402	0.4
Sydney	Fairfield	147 960	401	0.3
Sydney	South Sydney	55 840	406	0.7
Hunter	Newcastle	119 481	408	0.3
Mid-North Coast	Nambucca	14 529	414	2.8
North Western	Walgett	6 477	400	6.2
Total		457 746	2 431	0.5

^a Source: ABS estimated resident population data at 30 June 2003.

Geographic diversity

In order to get a picture of legal issues in socioeconomically disadvantaged areas across NSW, the LGAs were selected to include suburban areas within Sydney (Campbelltown, Fairfield and South Sydney), a major provincial centre (Newcastle) and rural/remote areas of NSW (Nambucca and Walgett). The LGAs were also chosen to cover geographically diverse areas of NSW. Given that the pilot testing was conducted in a southern area of NSW (the Bega Valley LGA), other areas of NSW were used for the main survey, including a northern coastal area (Nambucca) and an inland western area (Walgett).

Socioeconomic disadvantage

The risk score for cumulative socioeconomic disadvantage provided by Vinson (1999) was used to select LGAs in NSW that had relatively high levels of disadvantage. Vinson (1999) mapped postcodes in NSW (and Victoria) on the basis of their cumulative disadvantage risk score. This score is a composite score based on a range of socioeconomic indicators including the proportions of unemployed persons, low-income households, low-weight births, confirmed instances of child abuse, people who left school before 15 years of age, households receiving emergency assistance, convicted persons, child injuries, long-term unemployed persons and unskilled workers.¹⁰ It is important to note that cultural and linguistic diversity is not one of the indicators used.

The three selected non-Sydney LGAs all featured at least three postcodes among the 50 most disadvantaged postcodes in NSW according to the cumulative disadvantage risk score (five for Newcastle, four for Nambucca and three for Walgett). Of the 45 LGAs within Sydney, Campbelltown and South Sydney were the only two that featured postcodes among the top 50 disadvantaged postcodes for NSW. Fairfield was chosen as the third Sydney LGA because it had a relatively high cumulative disadvantage risk score and it has a culturally and linguistically diverse population (see below).

As already noted, although the present survey was conducted in disadvantaged areas, it should not be assumed that all the residents of those areas are disadvantaged.

Cultural and linguistic diversity

Indigenous status is often positively correlated with indicators of socioeconomic disadvantage (ABS 1995; Australian Institute of Health & Welfare (AIHW) 2005). Thus, it was considered important to include at least one LGA that has a high proportion of Indigenous Australians. According to the 2001 population census (ABS 2002b), Walgett LGA has a relatively large Indigenous population (21.5%) when compared with the NSW average (1.9%). Nambucca LGA also has a somewhat higher proportion of Indigenous Australians (5.4%) compared with the proportion for NSW overall.

Given that NSW also has a sizeable proportion of persons born outside Australia¹¹ and that ethnicity is often correlated with various socioeconomic indicators, it was decided to include at least one LGA that exhibits cultural and linguistic diversity. The Sydney LGA with the highest percentage of persons born outside Australia according to the 2001 population census is Fairfield (ABS 2002a). According to the census, a language other than English is spoken in 70.9 per cent of Fairfield households, with Vietnamese (15.5%), Chinese languages (10.2%) and Spanish (4.9%) being the most common non-English languages.

In order to maximise the opportunity of completing interviews with people from these three non-English language groups, the survey instrument was translated into Vietnamese, Cantonese and Spanish. Interviewers speaking these languages were made available for people who preferred to be interviewed in one of these languages.

Population size

To qualify for inclusion in the study, LGAs in rural or remote areas were also required to have adequate population sizes (at least 5000 persons).

Sampling of participants from disadvantaged areas

Random sampling using the electronic White Pages as at April 2003 was employed to select a pool of potential participants from each of the selected areas. In addition, quota controls were used to fulfil the following criteria:

• 400 residents aged 15 years or over from each LGA
• proportionate sampling in each LGA according to gender, age and postcode
• in Walgett LGA, proportionate sampling of Indigenous people
• in Fairfield LGA, proportionate sampling of the three cultural and linguistic groups of interest, namely Vietnamese, Chinese (Cantonese) and Spanish.

The objective of the sampling technique was to achieve a final sample that would reflect the demographic profile of the LGA populations.¹² Table 2.2 presents the gender and age breakdown for the overall sample and compares it to that for the population of the six LGAs combined. Tables B2a to B2f in Appendix B provide the corresponding tables for each LGA. According to the chi-square tests conducted, the proportions of males and females in the overall sample were not significantly different to those in the population. However, the overall age profile of the sample differed from that of the population, with the sample having a relatively lower proportion of 15 to 24 year olds (see notes for Table 2.2). The chi-square tests for each LGA revealed that there was no significant difference between the sample and population in the overall gender or age profiles. However, for five of the six LGAs, there was a significant difference between the sample and population in the distribution of males and females within age groups (see notes for Tables B2a to B2f in Appendix B).

Table 2.2: Gender and age breakdown in sample and population, all six LGAs, 2003

		Age (Years)
		15–24	25–34	35–44	45–54	55–64	65+	15–65+
Sample
Males	no. (%)	258 (10.6)	214 (8.8)	223 (9.2)	203 (8.4)	140 (5.8)	187 (7.7)	1 225 (50.5)
Females	no. (%)	145 (6)	249 (10.3)	258 (10.6)	247 (10.2)	159 (6.5)	145 (6)	1 203 (49.5)
Males and females	no. (%)	403 (16.6)	463 (19.1)	481 (19.8)	450 (18.5)	299 (12.3)	332 (13.7)	2 428 (100)
Population
Males	no. (%)	44 935 (9.8)	46 017 (10.1)	43 961 (9.6)	38 674 (8.4)	26 807 (5.9)	28 128 (6.1)	228 522 (49.9)
Females	no. (%)	43 864 (9.6)	44 942 (9.8)	42 173 (9.2)	38 001 (8.3)	24 935 (5.4)	35 309 (7.7)	229 224 (50.1)
Males and females	no. (%)	88 799 (19.4)	90 959 (19.9)	86 134 (18.8)	76 675 (16.8)	51 742 (11.3)	63 437 (13.9)	457 746 (100)

Notes:
1. Population data are estimated resident population as at 30 June 2003 (unpublished ABS data).
2. Each % is based on the cell no. divided by the total sample no. (2428) or the total estimated population (457 746), as appropriate. (Data on age were missing for three survey participants.)
3. Three chi-square tests were conducted comparing sample numbers with the corresponding expected numbers based on the population data. (The expected number for each cell = cell % for the population multiplied by the total sample no., e.g. expected no. of males 15–24 years = 9.8% x 2428 = 238).
a. one-way for gender: =0.25, df=1, p=0.616 (N=2431)
b. one-way for age: =18.75, df=5, p=0.002 (N=2428)
c. two-way for gender by age: =79.37, df=5, p=0.000 (N=2428)

In order to achieve 400 completed interviews for each LGA, a considerably larger pool of randomly selected telephone numbers was drawn from each postcode in the LGA. A larger pool was necessary to allow for inevitable wastage due to factors such as:

• inactive phone numbers (e.g. due to people moving or changing their numbers)
• inappropriate numbers (e.g. business rather than residential numbers)
• inability to achieve contact
• non-cooperation (e.g. outright refusals, terminations before interview completion, other forms of non-cooperation during interview)
• ineligibility according to the selection criteria
• quota requirements.

The number of randomly selected phone numbers comprising the pool for each LGA is presented in Table B3 in Appendix B. To achieve at least 2400 completed interviews overall (at least 400 from each LGA), calls were made to 24 725 (or 71.4%) of the total phone numbers in the pool (34 643).

Response rate

The minimisation of non-response is a quality control objective in any survey. However, what to report about response rate remains discretionary and there is no standard way of computing response rate (Biemer & Lyberg 2003; Groves 1989). Furthermore, it has been argued (e.g. Groves, Cialdini & Couper 1992) that the response rate is not as critical a measure of survey quality as is an understanding of the sociodemographic and behavioural differences between those who responded and those who refused. Unfortunately, given the voluntary nature of many surveys it is often difficult to obtain demographic and other details about persons who refused to participate or were not contactable, and it was not possible to obtain such details in the present survey.

The measures of survey response provided for the present study are the 'cooperation rate' and the 'response rate' as defined by Groves (Groves 1989; Groves, Biemer, Lyberg, Massey & Nicholls 1988; Groves et al.1992). These are commonly used measures of survey response and are based on the possible outcomes of phone calls from which an attempt was made to secure an interview.

The cooperation rate is I/(I+R) and the response rate is I/(I+R+NC+NI), where:

• I = number of completed interviews
• R = number of refused eligible units (including number of partial interviews)
• NC = number of non-contacted but eligible units
• NE = number of non-eligible units
• NI = number of other non-interviewed units.

The cooperation rate describes the success in persuading those who are contacted and eligible to complete an interview, and is based solely on those who are both eligible and contacted for the survey. The response rate also takes into account those from the sample pool who were eligible but not contacted.

The various outcomes of phone calls from which an attempt was made to secure an interview across the six LGAs are shown in Table 2.3. Tables B4a to B4f in Appendix B show the corresponding outcomes of calls for each LGA separately.

Table 2.3: Outcome of attempted phone contact, all six LGAs, 2003

Outcome	Groves's typology	No.		Examples of outcome
Interview completed	I	2 431
Refused a	-	7 469		Outright refusal (7097) Refused monitoring by a survey supervisor (159) Refused to complete interview (213)
Not eligible	NE	3 250		Failed to meet survey coverage criteria (2527) Business number (361) Language barrier b (362)
Not contacted and eligible	NC	847		Unable to determine appointment at call back
Not interviewed	NI	2 213		Contacted but surplus to quota needs
Not applicable	-	8 515		Phone number no longer exists (4383)No contact after 5 attempts c (4132)
Numbers called from pool		24 725
Numbers not called from pool		9 918
Total numbers in pool		34 643

^a R, the number of refusals who were eligible to participate, was unknown. There was no information on eligibility to participate for the outright refusals and those who refused monitoring by a survey supervisor.
^b Interviewer could not determine the language used by respondent.
^c On each attempt, either there was no answer after 10 rings, the phone was engaged, the call was answered by an answering machine or the number was dead.

The information about call outcomes for the present survey can generally be reconciled with Groves's definitions. The main exception is that R, the number of refused eligible units, was not known because eligibility to participate in the survey was unknown for the majority of those who refused. Of those who refused (7469), eligibility is known only for those who commenced but failed to complete an interview (213 or 2.9%). Other refusals occurred at the outset of the survey before screening questions on eligibility were asked.

As a result, the cooperation and response rates for the present survey need to be estimated. Assuming that all the refusals were eligible to participate¹³ gives a cooperation rate of 24.6 per cent and response rate of 18.8 per cent for the overall sample. However, these calculations are likely to underestimate survey response because some of the refusals are likely to have been ineligible. Excluding the refusals whose eligibility is unknown¹⁴ gives a cooperation rate of 91.9 per cent and a response rate of 42.6 per cent, which are likely to overestimate survey response.

It is possible to provide more precise estimates of the cooperation and response rates by estimating R. For example, if it is assumed that the proportion of eligible persons among those who refused is identical to the proportion of eligible persons among those who did not refuse, then R is 7469 x 0.628 = 4691.¹⁵ This estimate of R provides a cooperation rate of 34.1 per cent and a response rate of 23.9 per cent for the overall sample. Table B5 in Appendix B presents the estimates for the cooperation and response rates for each LGA.

Procedure

Pilot testing

The current survey was based on a pilot survey conducted in October and November 2002 with 306 residents of the Bega Valley LGA in south-east NSW. The LJF (2003) report provides a full description of the pilot survey, including its development, conduct and findings.

The pilot survey instrument was refined to produce the survey instrument for the present study. A detailed description of the changes to the pilot survey is presented in Table B6 in Appendix B. The main survey remains essentially very similar to the pilot survey. The major differences are that the main survey:

• includes three additional legal events (dispute with financial institution—question 22; local council problem—question 44D; domestic violence allegation against you—question 48)
• explicitly specifies the legally defined categories of discrimination (marital status, age, gender, religion, sexuality, ethnicity, disability) in the question on non-employment related discrimination (question 24)¹⁶
• asks about participants' three most recent legal events rather than about the most significant legal event in the last 12 months together with the other two most recent legal events¹⁷
• includes a question to measure the timing of the three most recent legal events (question 60).

Numerous minor changes were also made to improve clarity and ease of administration, including:

• minor changes to the wording of questions and pre-coded response categories
• removal of a few questions of limited import
• introduction of some further screening questions to assist with the interpretation of questions measuring legal events
• re-ordering of some questions
• replacement of some pre-coded responses with open-ended responses
• expanding the response category 'no answer provided' to include 'don't know'
• re-ordering the response categories so that 'question not applicable' and 'no answer provided/don't know' come after the other response categories.

Conduct of main survey

NCS Pearson, a social research firm, was engaged to conduct the interviews.¹⁸ To reduce cost and improve efficiency, the interviews were conducted by telephone. The survey was conducted between 23 August and 28 September 2003 by trained interviewers using Computer Assisted Telephone Interviewing (CATI) system software.¹⁹

Calls to potential participants were made between 5 pm and 9 pm on weekdays, and between 10 am and 6 pm on weekends, to maximise the likelihood of contacting persons across the demographic spectrum (including people who tend to be out during the day such as employed people and students). The call back policy adopted was to make five attempts to get through to an individual phone number, with call back attempts usually spaced out over a number of days and at least four hours apart. Once a given number was answered, if an eligible person was not at home, the policy was to make a further five call back attempts to talk to an eligible person within that household. Once contact was made with an eligible person who agreed to participate, the interview was conducted at that time if convenient. If inconvenient, an attempt was made to secure an agreement to conduct the interview at a specified time (a 'hard' appointment) or an agreement to conduct the interview at a later, unspecified time (a 'soft' appointment).

The average length of each interview was 21 minutes.

Advantages and limitations of the survey technique

Reporting of legal events

An advantage of the present survey is that it was not restricted to the measurement of civil legal events, but included criminal and family legal events.

The present study measured whether each of 101 different legal events was experienced by each individual, but did not measure the number of times each event was experienced by a given individual. For instance, a person who reported experiencing a problem related to renting accommodation may have experienced such a problem more than once during the 12-month reference period. Although this method of measurement would not have affected the estimation of the breadth of legal events experienced by each individual, it may have resulted in an under-counting of the total number of legal events experienced by some individuals and an underestimation of the total incidence rate. This limitation is shared with a number of the recent legal needs surveys.²⁰

As noted earlier, retrospective surveys such as the current one rely on participants' memories. The present study attempted to maximise the accurate recall of legal events through a number of strategies. Firstly, given that longer reference periods for recall are associated with increased risk of inaccurate or decayed memory of events (e.g. Biemer et al. 1991; Rubin 1982; Sudman & Bradburn 1973), the reference period was restricted to 12 months, which is considerably shorter than the reference periods of three to 5.5 years used in some other similar studies (e.g. Genn 1999; Genn & Paterson 2001; Maxwell et al. 1999; Pleasence et al. 2004b).²¹ Secondly, as already described, legal events were asked about in context and in considerable detail so that participants could readily identify whether they had experienced these events, without necessitating a particular level of legal knowledge.

Surveys are also subject to various forms of response bias, and rely on participants' willingness to reveal sensitive information (e.g. Biemer et al. 1991; Oppenheim 1992; Presser et al. 2004). Survey respondents sometimes provide socially desirable rather than honest answers in relation to highly personal matters, socially taboo issues, and socially disapproved or self-incriminating behaviour (Oppenheim 1992). For example, it has been observed that legal events of a highly personal nature, such as domestic violence, tend to be under-reported (Keys Young 1998). The present survey examined domestic violence as well as other potentially sensitive issues such as, for example, assault, criminal charges, child protection, discrimination and immigration. In order to minimise under-reporting of sensitive information, participants were reassured about the confidentiality and anonymity of the information provided. Furthermore, the use of telephone interviews, which provide greater anonymity than face-to-face interviews, may also have helped to minimise under-reporting of sensitive issues (Biemer et al. 1991; Oppenheim 1992). Nonetheless, the possibility of under-reporting sensitive information should be kept in mind when interpreting the results from the current survey.

Sample representativeness

The representativeness of a survey sample depends on both the sampling process and the survey response rate. The present survey was not strictly speaking a random probability sample of persons aged 15 years or over in the LGAs covered. Although the initial selection of potential participants involved random selection from the telephone directory, quota controls were used to obtain participants from demographic groups that tend to be relatively more difficult to find at home (e.g. younger people, men). These non-random elements of the sampling process reduce the generalisability of the results.

Furthermore, it should be noted that some demographic groups did not have the opportunity to be surveyed or were likely to be under-represented, such as:

1. persons living at residences without land-line telephones
2. persons who were homeless
3. persons who were institutionalised
4. persons who have difficulty completing long interviews (e.g. some persons with a mental or intellectual disability)
5. persons with poor English who were not provided with an interpreter (i.e. persons from language groups other than Vietnamese, Cantonese and Spanish).

It should also be remembered that the sample was drawn from six LGAs in NSW that exhibit disadvantage. The extent to which the results are generalisable to other disadvantaged areas of NSW is unclear.²²

The relatively low survey response rates (i.e. cooperation and response rates) also reduce the generalisability of the present results.²³ The finding of some differences between the gender and age profiles of the sample compared with those of the population also limits the generalisability of the results.

Given these limitations in terms of sample representativeness, the results from the present survey should be considered suggestive rather than conclusive.

Data analysis

NCS Pearson provided all the survey data to the LJF in de-identified form in a Statistical Package for the Social Sciences (SPSS) data file. Long verbatim responses were also provided in a Microsoft Excel spreadsheet.²⁴

Given that some participants reported experiencing more than one legal event in the reference period, two distinct units of analysis were used in the present study: a person-based approach and an event-based approach. The unit of analysis used in each case is specified in the tables, figures or the text in the results chapters.

Descriptive statistical analyses

Descriptive analyses, such as frequencies and percentages, were compiled from the survey data to address the first five aims of the study, that is:

1. the incidence of legal events
2. the response to legal events, including the use of legal services
3. the satisfaction with any assistance received
4. the resolution of legal events
5. the satisfaction with the outcome of legal events.

As outlined below, various types of inferential statistical analyses were used to address Aim 6, that is, the factors related to incidence, response, satisfaction with assistance, resolution and satisfaction with outcome.

Logistic regression analyses

The main inferential statistical analyses involved 15 logistic regression models, each examining the predictors of a different outcome variable. One model examined the predictors of reporting legal events of any type. A further 10 models examined the predictors of reporting each of the 10 most frequently occurring types of events.²⁵ The last four models examined the predictors of action taken for legal events, satisfaction with the assistance received for legal events, the resolution status of legal events and satisfaction with the outcome of legal events. The potential predictors examined for each outcome variable are detailed in Table B7 in Appendix B, and included various sociodemographic characteristics of the participants and various characteristics of the legal events.

Logistic regression is an appropriate form of multivariate analysis when the outcome variable is discrete rather than continuous. Like other forms of regression, it examines the relationship of an outcome variable to a set of potential predictor variables considered simultaneously. This technique determines the association of each potential predictor to the outcome variable when the effects of the other potential predictors are taken into account. That is, it determines the independent predictors of the outcome variable from the set of potential predictors examined (e.g. Agresti 1996; Hosmer & Lemeshow 2000; Menard 2002).

A separate logistic regression model was fitted for each outcome variable. All of the outcome variables were treated as binary variables, even though these variables are sometimes presented in the results chapters broken down into more than two categories in the corresponding cross-tabulations. In the regression analyses, all of the predictors were treated as categorical variables (i.e. variables with discrete categories). The set of potential predictors examined in each model is presented in Table B7 in Appendix B. The categories used for both the predictor and the outcome variables are also specified in Table B7.

Standard binary logistic regression was used for the first 11 outcome variables listed above. Mixed-effects binary logistic regression (Hedeker 1999, 2002) was used for the last four outcome variables. While standard logistic regression assumes the independence of observations, mixed-effects logistic regression allows for correlated observations. For the first 11 outcome variables, there was only one observation for each participant, so the observations were independent.²⁶ The last four outcome variables involved potentially correlated observations because some participants had multiple legal events. The mixed-effects regression appropriately adjusted for any correlation among events experienced by the same participant.

Significance of predictors

In each model, the overall significance of each potential predictor was examined at the 0.05 level. For significant predictors (e.g. age), further significance testing was conducted, involving comparisons between particular categories of the predictor (e.g. a comparison between 15 to 24 year olds and individuals 65 years or over).

More specifically, with the exception of the legal event group predictor, comparisons were made between one chosen category of each predictor (the reference category) and each other category of that predictor.

Basing comparisons on a single reference category is appropriate for predictors that only have a few categories and for predictors that have ordered categories (e.g. age, personal income, education). However, this method considerably limits the interpretation of nominal (non-ordered categorical) predictors that have numerous categories because many of these categories are not directly compared against each other.

Legal event group was the only nominal predictor with numerous categories in the present study. If the comparisons for legal event group had been based on a single reference category (i.e. a single legal event group), there would have been no comparisons between any of the remaining 14 legal event groups. As a result, comparisons of each legal event group were made against the average effect of all the legal event groups rather than against one specific legal event group (e.g. Menard 2002). Basing comparisons on the average effect allowed conclusions to be drawn about whether each legal event group was more or less likely than average to result in certain outcomes (e.g. seeking help).

For predictors that were significant overall, the significance of each comparison tested was also examined at the 0.05 level. Summary tables of the regression models are provided in the results chapters, while full details of the regressions are provided in Appendix C. The comparisons used for significant predictors are detailed in the summary tables and provided in the notes to the Appendix tables.

The odds ratio for each comparison is presented in the Appendix C tables. The odds ratio is a ratio of two sets of odds. For example, for the association between reporting legal events and gender, the odds ratio compares the odds of a male reporting legal events with the odds of a female reporting legal events.²⁷ An odds ratio that is not significantly different from the value of one (1.0) suggests that there is no real difference between these two sets of odds. An odds ratio that is significantly greater than one suggests that the first set of odds (for males) is higher than the set of odds for the reference category (for females). Conversely, an odds ratio that is significantly less than one suggests that the first set of odds is lower than the set of odds for the reference category. In the case of the legal event group predictor, the odds ratio compares the odds of a particular legal event group against the average odds for all legal event groups.

The 95 per cent confidence interval associated with each odds ratio is also presented in the Appendix C tables. The 95 per cent confidence interval provides, with 95 per cent certainty, the range of values that the odds ratio could take.

Further details about the logistic regression analyses performed are outlined in Appendix B.

Chi-square analyses

In addition to the logistic regression analyses, some chi-square analyses were also conducted to address aspects of Aim 6 that were not examined in the logistic regression analyses. For example, chi-square tests examined the relationship between:

• the incidence of the five least frequent legal event groups and the sociodemographic variables²⁸
• action taken in response to legal events and broad area of law
• region of residence and distance travelled to obtain assistance
• type of adviser and satisfaction with assistance
• method of resolution for resolved events and broad area of law
• method of resolution for resolved events and action taken in response to events
• satisfaction with outcome and satisfaction with assistance.

The chi-square test is a non-parametric test that examines whether there is a significant relationship between two or more categorical variables. The test is based on the cross-tabulation of the relevant variables, and compares the observed frequencies in each cell of the cross-tabulation with the frequencies expected if there were no relationship between the variables (e.g. Siegel & Castellan 1988). The chi-square test reveals the straightforward relationship between the two variables, when no other variables are taken into account (i.e. the bivariate relationship). The statistical significance of each chi-square test was examined at the 0.05 level.

Cluster analysis

The incidence patterns of legal events (Aim 1) were further examined via cluster analysis. Cluster analysis is an exploratory data analysis tool that groups observations according to their degree of relatedness (e.g. Aldenderfer & Blashfield 1984; Everitt, Landau & Leese 2001; SAS Institute Inc. 1993). Observations within a cluster are more closely related to one another than they are to observations in other clusters. In the present study, agglomerative hierarchical cluster analysis (AHCA) was used to identify which of the 15 legal event groups tended to be experienced by the same individuals (i.e. tended to co-occur) during the 12-month reference period.

AHCA starts with each observation (i.e. legal event group) in a separate cluster (i.e. 15 clusters). It then proceeds in a series of successive steps, with each step joining together the two clusters that are most similar into one cluster. In this way, legal event groups were combined into an increasingly smaller number of coherent clusters, until eventually, all event groups had been combined into one cluster.

Jaccard scores were used to measure the amount of similarity between legal event groups and centroid linkage was used as the clustering method.²⁹

The results of the AHCA were summarised in a hierarchical tree diagram, or dendrogram. The branches of the dendrogram illustrate which legal event groups were joined together at each step of the analysis. The length of the branches joining two or more legal event groups (as measured by the 'distance' shown on the x-axis of the dendrogram) indicate the degree of similarity between those legal event groups. More specifically, the shorter the length of the branches joining legal event groups, the greater the similarity (or co-occurrence) of those event groups, and the earlier in the analysis that these event groups were combined into one cluster.

The number of clusters formed by a particular stage in the analysis can be made evident graphically by 'cutting' (i.e. drawing a line through) the dendrogram at the distance corresponding to that stage, and noting which clusters were formed below that distance. There is no single established method for deciding the 'best cut' of the dendrogram, that is, for deciding on the optimal number of clusters that best describes the relationships between observations. The various formal tests available for this purpose often provide different results and, consequently, heuristic approaches are commonly used (Aldenderfer & Blashfield 1984; Everitt et al. 2001; SAS Institute Inc. 1993). The most basic heuristic approach is to 'cut' the dendrogram according to the subjective inspection of the different levels of the tree. A common method used to assist in determining the best cut involves examining the distance between the fusion coefficients at each stage, and cutting the dendrogram at a relatively large jump in the value of the coefficient (Aldenderfer & Blashfield 1984; Everitt et al. 2001). In the present case, the optimal number of clusters was determined using a combination of subjective inspection and the change in the fusion coefficient.

Factor analysis

The incidence patterns of legal events (Aim 1) were also examined via factor analysis. Factor analysis is a data reduction technique used to identify underlying dimensions, or factors, that explain the pattern of relationships within a set of observed variables (e.g. Green & Salkind 2003; Tabachnick & Fidell 2001). In this instance, exploratory factor analysis was carried out to examine the relationships between the 15 legal event groups and to identify underlying dimensions. Legal event groups that tend to co-occur will contribute to the same underlying dimension or factor, while unrelated legal event groups will contribute to different factors. Legal event groups are considered to contribute to a particular underlying dimension or factor if they 'load' significantly on that factor.

Factor analysis generally involves two stages: factor extraction and factor rotation (Green & Salkind 2003). Principal component extraction was used and the number of factors extracted was based on an examination of a scree plot displaying eigenvalues. Three factors were identified. Varimax (orthogonal) rotation with Kaiser normalisation was used to maximise the loading of each variable on a single factor, while minimising the loadings of the variable on the other factors. Legal event groups with loadings of 0.320 or higher on a given factor were considered to load significantly on that factor (e.g. Tabachnick & Fidell 2001).

Missing values

The number of missing values for each descriptive and inferential statistical analysis is provided in the table notes. The numbers of participants and/or events included in each logistic regression analysis are also listed in Table B7 in Appendix B. Each analysis was based only on participants and/or events that had data on all the variables of interest used in the analysis (i.e. listwise deletion of missing values was used).