Effect of Indigenous status on sentence outcomes for serious assault offences

Abstract

Recent studies have found evidence of a racial bias in sentencing in Australia despite previous work showing the opposite. However, in some cases, these studies fail to adjust properly for differences in offence seriousness and in others fail to control for important covariates. Our aim in this study was to investigate the relationship between Indigenous status and likelihood of imprisonment for serious assault offences, matching on offence seriousness and controlling for additional covariates. Multivariate logistic regression with generalized estimating equations was used to explore the effect of Indigenous status on the likelihood of a prison sentence while controlling for the other sentence-relevant factors. Interaction effects were examined by fitting a logistic regression model with all covariates separately for Indigenous and non-Indigenous offenders and comparing the coefficient of each covariate for differences. The results show that the marginal increase in the risk of a prison sentence for Indigenous offenders is less than one percentage point (3.40% for Indigenous offenders compared with 2.45% for non-Indigenous). Significant interaction effects between Indigenous status and whether or not the offence was domestic violence related (p = .024), whether or not the offence was committed after 2011 (p = .020) and whether or not the offender had previously been issue a community service order for a serious assault offence (p = .025) were found, each of which altered the probability of imprisonment. Despite this, no interaction effect was found that resulted in a difference in marginal probability of imprisonment greater than 1.2 percentage points between Indigenous and non-Indigenous offenders. We conclude that while there is a significant effect of Indigenous status on sentencing, the effect is vastly reduced when controlling for additional covariates.

Keywords

Disparity generalized estimating equations imprisonment Indigenous status sentencing serious assault

Introduction

One of the main problems currently faced by the criminal justice systems in Australia is the gross over-representation of Indigenous offenders in the prison population. In September 2016, nearly 25% of all adult prisoners in New South Wales (NSW) identified as either Aboriginal or Torres Strait Islander status (NSW Bureau of Crime Statistics and Research [BOCSAR], 2016), despite only comprising 2.9% of the general population (June 2011 estimate, Australian Bureau of Statistics, 2013).

One possible explanation for this discrepancy is systemic bias in the response of the criminal justice system to Indigenous defendants and offenders, as claimed by Anthony (2013); Blagg, Morgan, Cunneen, and Ferrante (2005); and Cunneen, Allison, and Schwartz (2014). However, research seems to be mixed regarding whether or not a systemic bias exists at sentencing. While there is qualitative research to suggest that a bias may exist in sentencing (Anthony, 2013; Anthony, Bartels, & Hopkins, 2015; Baldry & Cunneen, 2014; Cunneen, 1992), empirical results on the matter have been mixed.

In Australia, studies by Bond and Jeffries (2010); Bond and Jeffries (2011a); Bond and Jeffries (2011b); Jeffries and Bond (2009); Snowball and Weatherburn (2006); Snowball and Weatherburn (2007) found little or no evidence of a bias between Indigenous and non-Indigenous offenders at sentencing. In fact, in their review of the quantitative literature, Jeffries and Bond (2012) concluded:

Once crucial sentencing factors are held constant (especially, current and past offending), sentencing outcomes for Indigenous and non-Indigenous offenders either achieve parity or the gap [in sentencing outcomes] is considerably reduced … In circumstances where disparity remains, there is evidence to suggest that, Indigenous defendants are at times treated leniently in comparison with their non-Indigenous counterparts. (p. 23)

Since Jeffries and Bond (2012) published their review, three studies have emerged once again raising the possibility of racial bias in the sentencing of Indigenous offenders. The first (Donnelly & Poynton, 2015) was concerned with the question of whether domestic violence offenders were dealt with more leniently by the courts than offenders guilty of similar offences that were non-domestic. One of the covariates used in the study was Indigenous status, which was found to have a significant and substantial positive effect on the risk of a prison sentence even after controlling for a wide range of other factors. The odds ratio associated with Indigenous status was found to be 1.37, meaning that all else being equal, the odds of an Indigenous offender receiving a prison sentence were 37% higher than that of a non-Indigenous offender. This finding echoes the results of a similar study carried out in the same year by Jeffries and Bond (2015). They found that the risk of imprisonment for Indigenous offenders was the same, regardless of whether they were convicted of a domestic violence or a non-domestic violence offence. Where the offender was non-Indigenous, however, the risk of imprisonment was substantially lower if the offence involved domestic violence than if it did not involve such violence.

Lockwood, Hart, and Stewart (2015) also report mixed results in sentencing disparities between Indigenous and non-Indigenous offenders in Queensland courts, using conjunctive analysis of case configurations (CACC) to match appropriate observations. CACC is a technique which matches observations on all observable covariates except for the variable of interest (Indigenous status, in this case). The technique was used in a criminological context by Miethe, Hart, and Regoeczi (2008) who used it to examine factors influencing the sentencing of federal drug offenders in the United States. Lockwood et al. (2015) used it to compare sentencing outcomes between Indigenous and non-Indigenous offenders who only differed only in terms of Indigenous status. They found that the difference in imprisonment rate between the two types of offenders ranged between −35 percentage points and 44 percentage points in the lower court and between −30 percentage points and 56 percentage points in the higher court. A negative difference means that Indigenous offenders were sentenced less harshly, whereas a positive difference means they were sentenced more harshly.

One likely explanation for the inconsistent results found in quantitative studies of racial bias in sentencing is that different studies employ different controls. Studies of racial bias in sentencing need to take particular care to include variables that have been identified by the appeal courts or the legislature as relevant to sentencing. Drug and alcohol abuse, for example, has been identified by appeal courts as a relevant and sometimes aggravating consideration in the sentencing for violent offences (see NSW Sentencing Council, 2009, p. 46), yet neither Donnelly and Poynton (2015) nor Jeffries and Bond (2015) included controls for this factor. Again, past studies have found that prior penalties exert an effect on penalty selection over and above the effect of prior convictions (see, for example, Snowball, 2008). Courts also seem more likely to impose prison sentences where an offender has previously breached an apprehended violence order (Donnelly and Poynton, 2015). Jeffries and Bond (2015) included no controls for prior penalties or prior breaches of orders. Some or all of the effects of Indigenous status on the risk of imprisonment might be due to failure to control for these factors. For example, when Bond and Jeffries (2009) and Snowball and Weatherburn (2007) included controls covering (among other things) prior criminal history (and in the case of Bond and Jeffries, substance abuse), they showed that as more controls were added to the model, the odds ratio for the effect of Indigenous status on imprisonment decreased.

Another important procedural requirement in studies examining the effect of racial bias in sentencing is to avoid grouping offences with very different statutory maxima. The statutory maximum penalty is an important marker of the seriousness with which the legislature views a particular offence. Offenders convicted of offences with very different statutory maxima are likely to receive very different sentences. Failure to control for this could bias estimates of the effect of Indigenous status on the risk of imprisonment. Two offenders convicted of assault who differ only in their Indigenous status may receive different penalties because one has committed a more serious form of assault (e.g. assault occasioning grievous bodily harm) than the other (e.g. common assault). Jeffries and Bond (2015) endeavoured to mitigate this risk by including a measure of offence seriousness using the National Offence Index (Australian Bureau of Statistics, 2009), but it is doubtful that this strategy succeeded. For one thing, their sample consisted of all adult offenders convicted of a ‘personal’ offence. This would have included offenders convicted of offences as diverse (in terms of seriousness) as assault, sexual assault and robbery. For another, the National Offence Index also conflates offences of widely varying seriousness. The index, for example, makes no distinction between assault occasioning actual bodily harm (assault ABH) and assault occasioning grievous bodily harm (assault GBH) although assault ABH has a maximum penalty of 5 years imprisonment and assault GBH has a maximum penalty of 25 years imprisonment.

The purpose of this study is to revisit the questions addressed by Donnelly and Poynton (2015) and Jeffries and Bond (2015) using (a) a more extensive set of controls and (b) assault offences matched in terms of seriousness. More specifically, in addition to the usual demographic and prior record controls, we include controls for drug and alcohol abuse, prior penalties (prior suspended sentence, prior prison sentence) and prior breach of an apprehended violence order. We also include specific controls for the seriousness of the assault, as indicated by the exact nature of the charge (conviction for assault occasioning ABH vs. conviction for assault occasioning GBH).

Method

Data sources

Data were taken from BOCSAR’s Re-Offending Database. The sample we considered covered offenders who were found guilty of a serious assault offence (ANZSOC code 0211) as their principal offence in a NSW Local Court between January 2009 and December 2014 and who received a prison sentence, suspended sentence or suspended supervised sentence only. The sample was limited to those offenders whose principal offence was an assault involving ABH or GBH (the law part codes for these offences can be found in Table 6 of Appendix 1). The law part codes used only took effect from 31 March 2008, so offenders were restricted to those whose principal offence had occurred after this date.

Explanatory covariates

The majority of covariates included as explanatory variables were those used by Donnelly and Poynton (2015). These variables were as follows:

Age: The age of the offender at the time of index finalisation. This was classified into categories of 18–24 years; 25–34 years; 35–44 years; 45+ years

Gender: The gender of the offender

Indigenous status: The Indigenous status of the offender as recorded at the index contact, classified as non-Indigenous/unknown; Indigenous

Year of Finalisation: The year the index matter was finalised, classified into categories of 2009–2011; 2012–2014

Plea: How the offender pleaded to the index offence, classified as not guilty; guilty/other

Concurrent offences: Number of proven concurrent offences other than the principal offence at the index contact, classified as none; one; two; three or more

Severity: The severity of assault, classified as ABH; GBH (classified using law part codes)

DV Related: Whether or not the principal offence was a domestic violence related offence, classified as DV related; non-DV related (classified using law part codes)

Prior proven court appearances: Number of finalised court appearances with proven offences in the five years prior to the index appearance, classified as none; one; two; three or more

Prior proven serious assaults: Same as prior proven court appearances, restricted to serious assault offences (ANZSOC codes 211 and 212), classified as none; one or more

Prior prison sentences: Same as prior proven court appearances, restricted to appearances where the offender received a prison sentence for the principal offence, classified as none; one; two or more

Prior prison sentences for serious assault: Same as prior prison sentences, restricted to appearances where the prison sentence was for serious assault offences (ANZSOC codes 211 and 212)

Prior suspended prison sentences: Same as prior prison sentences, but for court appearances where the offender received a suspended sentence rather than a prison sentence, classified as none; one or more

Prior proven breaches of violence/non-violence orders: Number of finalised court appearances with proven breaches of violence/non-violence orders in the five years prior to the index appearance, classified as none; one or more

In addition, the Level of Service Inventory-Revised (LSI-R) domain 8 score was included as an indicator for previous alcohol/drug abuse. The LSI-R score was treated as a categorical variable, with an additional category for offenders who had no LSI-R score present. Because the decision on what sentence to impose depends partly on how the defendant has responded to previous penalties, a number of variables looking at the prior penalties received for serious assault offences were also included. These were as follows:

Prior suspended prison sentences for serious assault offences: Same as prior suspended sentences, but for court appearances where the suspended sentence was for a serious assault offence (ANZSOC codes 211 and 212), classified as none; one or more

Prior supervised suspended prison sentences for serious assault offences: Same as prior suspended sentences for serious assault, but for court appearances where the suspended sentence was supervised, classified as none; one or more

Prior community service orders for serious assault offences: Same as prior suspended sentences for serious assault, but for court appearances where a community service order was imposed rather than a suspended sentence, classified as none; one or more

Prior good behaviour bonds issued for serious assault offences: Same as prior suspended sentences for serious assault, but for court appearances where a good behaviour bond was issued rather than a suspended sentence, classified as none; one or more

Prior supervised good behaviour bonds issued for serious assault offences: Same as prior suspended sentences for serious assault, but for court appearances where a supervised good behaviour bond was issued rather than a suspended sentence, classified as none; one or more

Statistical analysis

Multivariate logistic regression with generalized estimating equations (GEEs) was used to analyse the data. We attempted to perform CACC, but found that the cases were spread too evenly among the configurations to give us accurate estimates, as only 18 of the 532 case configurations in our data set containing at least one Indigenous and one non-Indigenous offender described 10 or more cases (a suggested minimum frequency for CACC). Multivariate logistic regression was chosen as this technique handles binomial response variables (our response variable was a binary indicator of whether or not a particular offender received a prison sentence) and it allows one to separate out and see the effect of different covariates on the response variable. GEE was used to fit this model as this corrects for intra-class correlations, which could be present within particular courts. To illustrate the effect of the covariates on the change in odds of imprisonment between Indigenous and non-Indigenous offenders, the model will first be fitted on only demographic variables for the offenders, before other variables are gradually added to the model. The change in the coefficient for Indigenous status will be examined as more variables are added to the model. The variables to be used in each model are shown in Table 1.

Table 1.

Descriptions of the different models used in this analysis.

Model name	Variables
Demographics model	(Intercept)
	25–34 years vs. 18–24 years
	35–44 years vs. 18–24 years
	45 and older vs. 18–24 years
	Male vs. female
Index offence model	All variables in the demographics model as well as
	Domestic violence related vs. not
	GBH vs. ABH
	1 Concurrent offence vs. 0
	2 Concurrent offences vs. 0
	3 or more concurrent offences vs. 0
	2012–2014 vs. 2009–2011
	Not guilty plea vs. guilty/other
Prior appearances model	All variables in the plea model as well as
	1 prior court appearance with a proven offence in the past five years vs. 0
	2 prior court appearances with a proven offence in the past five years vs. 0
	3 or more prior court appearances with a proven offence in the past five years vs. 0
	1 or more prior court for serious assault in the past five years vs. 0
	LSIR dom 8 score 1 vs. 0
	LSIR dom 8 score 2 vs. 0
	LSIR dom 8 score 3 vs. 0
	LSIR dom 8 score 4 vs. 0
	LSIR dom 8 score 5 vs. 0
	LSIR dom 8 score 6 vs. 0
	LSIR dom 8 score 7 vs. 0
	LSIR dom 8 score 8 vs. 0
	LSIR dom 8 score 9 vs. 0
	LSIR dom 8 score missing vs. 0
	1 prior prison sentence in the past five years vs. 0
	2 or more prior prison sentence in the past five years vs. 0
	1 or more prior prison sentence in the past five years for serious assault vs. 0
	1 or more prior suspended sentences in the past five years
	1 or more prior breaches of violence orders in the past five years
	1 or more prior suspended sentences for serious assault in the past five years
	1 or more prior supervised suspended sentences for serious assault in the past five years
	1 or more prior community service orders for serious assault in the past five years
	1 or more prior good behaviour bonds for serious assault in the past five years
	1 or more prior supervised good behaviour bonds for serious assault in the past five years

Due to the large number of variables, interaction effects were not included in the binomial GEE regression. Possible interaction effects were examined by fitting the model separately for Indigenous and non-Indigenous offenders, then comparing the coefficient estimates for each parameter for the two models. Standard multivariate logistic regression was used rather than GEE regression at this stage, as it tests for significance between the coefficients of the two models. The coefficients were tested for difference using the procedure outlined by Paternoster, Brame, Mazerolle, and Piquero (1998).

Results

Offender characteristics

Table 2 shows the demographic characteristics of the offenders, including the number of offenders in each category for each variable, as well as the percentage of offenders in each category who received a prison sentence and the significance of the difference between the different categories¹. It can be seen that 55% of Indigenous offenders received a prison sentence, compared with 36% of non-Indigenous offenders. The age group of 25–34 years had the highest imprisonment risk of all age groups (45%), and male offenders were more than twice as likely to be imprisoned as female offenders (44% vs. 19%). Increases in the number of concurrent or prior offences (of any type) increased the risk of imprisonment. Prior court appearances where the offender received a prison sentence, suspended sentence, good behaviour bond (supervised or otherwise), or community service order also increased the risk of imprisonment (whether or not the previous principal offence was a serious assault offence). Similar rates of imprisonment can be seen regardless of whether the offence was domestic violence related, and regardless of whether the offence occurred in the period 2009–2011 or 2012–2014.

Table 2.

Demographic characteristics of serious assault offenders.

Variable	Categories	Number	% Prison
Indigenous status*	Non-Indigenous/ unknown	6948	35.6
Indigenous status*	Indigenous	2259	55.2
Age*	18–24 years	2477	38.1
	25–34 years	3118	45.3
	35–44 years	2455	41.2
	45 years or older	1157	30.8
Gender*	Female	1233	19.2
Gender*	Male	7974	43.7
DV related	No	4627	41.4
DV related	Yes	4580	39.5
Severity	ABH	8897	39.7
Severity	GBH	310	62.9
Number of concurrent offences*	0	3473	26.5
	1	2144	36.4
	2	1268	46.1
	3 or more	2322	62.0
Year of offence	2009–2011	5106	40.8
Year of offence	2012–2014	4101	40.1
Plea^*	Guilty/other	7828	38.7
Plea^*	Not guilty	1379	50.1
Number of prior court appearances in the past five years with a proven offence*	0	2311	20.2
	1	1936	28.3
	2	1522	38.4
	3 or more	3438	61.8
Number of prior court appearances in the past five years with a proven serious assault offence*	0	7171	34.2
	1 or more	2036	62.4
LSIR domain 8 score*	0	578	13.0
	1	227	23.8
	2	316	22.5
	3	534	22.3
	4	991	23.7
	5	1131	36.3
	6	1201	42.6
	7	976	52.3
	8	565	55.0
	9	218	70.6
	LSIR score missing	2470	51.5
Number of prior prison sentences imposed in the past five years*	0	6864	28.0
	1	1098	68.9
	2 or more	1245	84.3
Number of prior prison sentences imposed in the past five years for a serious assault offence*	0	8268	35.7
	1 or more	939	82.2
Number of prior suspended sentences imposed in the past five years*	0	7731	35.8
	1 or more	1476	64.6
Number of breaches of violence orders issued in the past five years*	0	7507	35.0
	1 or more	1700	64.4
Number of prior suspended sentences imposed in the past five years for serious assault offences*	0	8828	39.3
	1 or more	379	68.1
Number of prior suspended sentences imposed in the past five years for serious assault offences*	0	8930	39.6
	1 or more	277	67.9
Number of prior community service orders imposed in the past five years for serious assault offences*	0	9021	40.1
	1 or more	186	57.0
Number of prior good behaviour bonds imposed in the past five years for serious assault offences*	0	8325	39.1
	1 or more	882	52.7
Number of prior supervised good behaviour bonds imposed in the past five years for serious assault offences*	0	8650	39.3
	1 or more	557	58.0

Significant association between the variable and probability of imprisonment.

Table 3 shows the demographic characteristics split by Indigenous status. The majority of both Indigenous and non-Indigenous offenders were male, aged between 25 and 34 years of age, had committed an ABH offence as their principal offence and had no concurrent offences. The larger differences between the two groups occur for variables relating to prior criminal history and previous alcohol/drug abuse (LSI-R domain 8 score). Indigenous offenders tended to have more prior proven offences (both overall and specifically for serious assault) as well as more prior proven offences for which a prison sentence, suspended sentence, good behaviour bond (supervised or otherwise), or community service order was imposed. For example 52% of Indigenous offenders had three or more prior court appearances with a proven offence, compared to 33% of non-Indigenous offenders. Also, Indigenous offenders are more likely to have higher LSI-R domain 8 scores; with 54% of Indigenous offenders having LSI-R scores of 5–9, compared with 41% of non-Indigenous offenders. A similar proportion of offenders in each group had missing LSI-R scores (26% for non-Indigenous offenders and 28% for Indigenous).

Table 3.

Demographic characteristics of offenders by Indigenous status.

Variable	Categories	Non-Indigenous		Indigenous		Total
Variable	Categories	Number	%	Number	%	Number	%
Prison sentence	Yes	4472	64.4	1011	44.8	5483	59.6
Prison sentence	No	2476	35.6	1248	55.2	3724	40.4
Age	18–24 years	1780	25.6	697	30.9	2477	26.9
	25–34 years	2328	33.5	790	35.0	3118	33.9
	35–44 years	1874	27.0	581	25.7	2455	26.7
	45 years or older	966	13.9	191	8.5	1157	12.6
Gender	Female	814	11.7	419	18.5	1233	13.4
Gender	Male	6134	88.3	1840	81.5	7974	86.6
DV related	No	3532	50.8	1095	48.5	4627	50.3
DV related	Yes	3416	49.2	1164	51.5	4580	49.7
Severity	ABH	6699	96.4	2198	97.3	8897	96.6
Severity	GBH	249	3.6	61	2.7	310	3.4
Number of concurrent offences	0	2787	40.1	686	30.4	3473	37.7
	1	1649	23.7	495	21.9	2144	23.3
	2	927	13.3	341	15.1	1268	13.8
	3 or more	1585	22.8	737	32.6	2322	25.2
Year of offence	2009–2011	3833	55.2	1273	56.4	5106	55.5
Year of offence	2012–2014	3115	44.8	986	43.6	4101	44.5
Plea	Guilty/other	5909	85.0	1919	84.9	7828	85.0
Plea	Not guilty	1039	15.0	340	15.1	1379	15.0
Number of prior court appearances in the past five years with a proven offence	0	2012	29.0	299	13.2	2311	25.1
	1	1543	22.2	393	17.4	1936	21.0
	2	1120	16.1	402	17.8	1522	16.5
	3 or more	2273	32.7	1165	51.6	3438	37.3
Number of prior court appearances in the past five years with a proven serious assault offence	0	5676	81.7	1495	66.2	7171	77.9
	1 or more	1272	18.3	764	33.8	2036	22.1
LSIR domain 8 score	0	534	7.7	44	1.9	578	6.3
	1	203	2.9	24	1.1	227	2.5
	2	269	3.9	47	2.1	316	3.4
	3	455	6.5	79	3.5	534	5.8
	4	794	11.4	197	8.7	991	10.8
	5	810	11.7	321	14.2	1131	12.3
	6	864	12.4	337	14.9	1201	13.0
	7	650	9.4	326	14.4	976	10.6
	8	395	5.7	170	7.5	565	6.1
	9	149	2.1	69	3.1	218	2.4
	LSIR score missing	1825	26.3	645	28.6	2470	26.8
Number of prior prison sentences imposed in the past five years	0	5524	79.5	1340	59.3	6864	74.6
	1	709	10.2	389	17.2	1098	11.9
	2 or more	715	10.3	530	23.5	1245	13.5
Number of prior prison sentences imposed in the past five years for a serious assault offence	0	6443	92.7	1825	80.8	8268	89.8
	1 or more	505	7.3	434	19.2	939	10.2
Number of prior suspended sentences imposed in the past five years	0	5968	85.9	1763	78.0	7731	84.0
	1 or more	980	14.1	496	22.0	1476	16.0
Number of breaches of violence orders issued in the past five years	0	5872	84.5	1635	72.4	7507	81.5
	1 or more	1076	15.5	624	27.6	1700	18.5
Number of prior suspended sentences imposed in the past five years for serious assault offences	0	6716	96.7	2112	93.5	8828	95.9
	1 or more	232	3.3	147	6.5	379	4.1
Number of prior supervised suspended sentences imposed in the past five years for serious assault offences	0	6782	97.6	2148	95.1	8930	97.0
	1 or more	166	2.4	111	4.9	277	3.0
Number of prior community service orders imposed in the past five years for serious assault offences	0	6833	98.3	2188	96.9	9021	98.0
	1 or more	115	1.7	71	3.1	186	2.0
Number of prior good behaviour bonds imposed in the past five years for serious assault offences	0	6369	91.7	1956	86.6	8325	90.4
	1 or more	579	8.3	303	13.4	882	9.6
Number of prior supervised good behaviour bonds imposed in the past five years for serious assault offences	0	6601	95.0	2049	90.7	8650	94.0
	1 or more	347	5.0	210	9.3	557	6.0
Total		6948	100.0	2259	100.0	9207	100.0

Multivariate models

Table 4 shows the regression coefficient estimates and the standard errors for the different variables for each of the models described in Table 1.

Table 4.

Coefficient estimates and standard errors for variables in each model.

Variable	Demographics model	Index offence model	Prior penalties model
(Intercept)	−2.05 (0.12)*	−2.61 (0.13)*	−3.68 (0.21)*
Indigenous vs. non-Indigenous/ unknown	0.82 (0.06)*	0.75 (0.06)*	0.34 (0.07)*
25–34 years vs. 18–24 years	0.28 (0.06)*	0.33 (0.06)*	0.26 (0.06)*
35–44 years vs. 18–24 years	0.16 (0.06)*	0.28 (0.06)*	0.27 (0.07)*
45 years or older vs. 18–24 years	−0.23 (0.08)*	−0.06 (0.07)	0.07 (0.07)
Male vs. female	1.30 (0.09)*	1.31 (0.08)*	1.18 (0.08)*
DV related vs. non-DV related		−0.21 (0.05)*	−0.19 (0.05)*
GBH vs. ABH		1.18 (0.12)*	1.55 (0.14)*
One concurrent offence vs. none		0.45 (0.06)*	0.41 (0.06)*
Two concurrent offences vs. none		0.79 (0.08)*	0.71 (0.09)*
Three or more concurrent offences vs. none		1.44 (0.08)*	1.39 (0.08)*
Offence occurred 2012–2014 vs. offence occurred 2009–2011		−0.06 (0.06)	0.12 (0.07)
Not guilty plea vs. guilty plea/other		0.51 (0.07)*	0.38 (0.07)*
1+ prior proven offence in the past five years vs. 0			0.03 (0.08)
2 prior proven offences in the past five years vs. 0			0.16 (0.09)
3+ prior proven offences in the past five years vs. 0			0.37 (0.08)*
1 + prior proven serious assault offences in the past five years vs. 0			0.14 (0.11)
LSIR domain 8 score of 1 vs. 0			0.46 (0.19)*
LSIR domain 8 score of 2 vs. 0			0.19 (0.18)
LSIR domain 8 score of 3 vs. 0			0.15 (0.21)
LSIR domain 8 score of 4 vs. 0			0.14 (0.16)
LSIR domain 8 score of 5 vs. 0			0.42 (0.15)*
LSIR domain 8 score of 6 vs. 0			0.43 (0.15)*
LSIR domain 8 score of 7 vs. 0			0.62 (0.15)*
LSIR domain 8 score of 8 vs. 0			0.64 (0.16)*
LSIR domain 8 score of 9 vs. 0			1.52 (0.25)*
Missing LSIR domain 8 score vs. 0			1.16 (0.15)*
1+ prior prison sentence in the past five years vs. 0			1.09 (0.12)*
2+ prior prison sentences in the past five years vs. 0			1.71 (0.11)*
1+ prior prison sentences for serious assault vs. 0			0.26 (0.14)
1+ prior suspended sentences vs. 0			0.45 (0.08)*
1+ prior breaches of violence orders vs. 0			0.28 (0.07)*
1+ prior suspended sentences for serious assault offences vs. 0			0.18 (0.25)
1+ prior supervised suspended sentences for serious assault vs. 0			−0.15 (0.24)
1+ prior community service orders for serious assault offences vs.0			0.07 (0.17)
1+ prior good behaviour bonds for serious assault vs. 0			0 (0.17)
1+ prior prison sentences for serious assault vs. 0			0.17 (0.15)
Intra-court correlation²	0.08 (0.04)	0.06 (0.03)	0.04 (0.02)

Significant at the 5% level.

The coefficient for Indigenous status is both positive and significant in all three models, showing that no matter what variables are controlled for, Indigenous offenders (all else being equal) have an increased likelihood of receiving a prison sentence. However, the size of the coefficient decreases as more variables are added to the model. This shows that the difference in the likelihood of an Indigenous and a non-Indigenous offender receiving a prison sentence decreases as more variables are controlled for. To illustrate this, we calculated the marginal probabilities of receiving a prison sentence for the referent-level offenders for each model for both Indigenous and non-Indigenous offenders, shown in Figure 1.

Figure 1.

Marginal probabilities of offenders by Indigenous status.

Figure 1 shows the marginal probability of receiving a prison sentence for an Indigenous and non-Indigenous offender of the referent level for all other covariate variables (a female offender, aged 18–24 years, found guilty of an ABH, non-DV offence with no concurrent or prior offences of any kind, and an LSI-R score of 0). The probabilities shown in Figure 1 support the conclusion drawn from the coefficient estimates in Table 4 – the more covariates that can be controlled for, the smaller the difference in probability of imprisonment. Once all covariates have been controlled for, the difference in probability is less than one percentage point (as opposed to approximately 11 percentage points for the demographics only model).

Our assertions about the marginal probabilities assume that there are no interaction effects present. In order to test this, the full model was fit to both Indigenous and non-Indigenous offenders, and the value of each coefficient in each model was compared. In order to make statistical inferences between the coefficients in each model, logistic regression needed to be used rather than GEE regression. While regular logistic regression cannot account for intra-court correlation, the use of logistic regression (and hence ignoring the possibility of intra-court correlation) was justified given that the intra-court correlation found when fitting the prior appearances model was 0.04 (almost negligible). The results of this comparison are shown in Table 5.

Table 5.

Comparison of coefficients for Indigenous and non-Indigenous models.

Variable	Coefficients		Difference
Variable	Non-Indigenous model	Indigenous model	Difference
(Intercept)	−3.83	−3.53	−0.30
25–34 years vs. 18–24 years	0.27	0.32	−0.05
35–44 years vs. 18–24 years	0.32	0.14	0.17
45 years or older vs. 18–24 years	0.03	0.30	−0.28
Male vs. female	1.11	1.43	−0.32
DV related vs. Non-DV related	−0.26	0.02	−0.28*
GBH vs. ABH	1.58	1.95	−0.37
One concurrent offence vs. none	0.45	0.43	0.02
Two concurrent offences vs. none	0.76	0.81	−0.05
Three or more concurrent offences vs. none	1.48	1.53	−0.05
Offence occurred 2012–2014 vs. offence occurred 2009–2011	0.18	−0.11	0.29*
Not guilty plea vs. guilty plea/other	0.38	0.46	−0.08
1 prior proven offence in the past five years vs. 0	0.09	−0.10	0.19
2 prior proven offences in the past five years vs. 0	0.14	0.09	0.05
3+ prior proven offences in the past five years vs. 0	0.33	0.44	−0.11
1 + prior proven serious assault offences in the past five years vs. 0	−0.03	0.38	−0.42
LSI-R domain 8 score of 1 vs. 0	0.54	0.46	0.08
LSIR domain 8 score of 2 vs. 0	0.33	−0.10	0.42
LSIR domain 8 score of 3 vs. 0	0.17	0.38	−0.21
LSIR domain 8 score of 4 vs. 0	0.23	−0.12	0.35
LSIR domain 8 score of 5 vs. 0	0.39	0.62	−0.23
LSIR domain 8 score of 6 vs. 0	0.44	0.56	−0.12
LSIR domain 8 score of 7 vs. 0	0.75	0.58	0.17
LSIR domain 8 score of 8 vs. 0	0.68	0.85	−0.17
LSIR domain 8 score of 9 vs. 0	1.66	1.65	0.01
Missing LSIR domain 8 score vs. 0	1.30	1.05	0.25
1 prior prison sentence in the past five years vs. 0	1.32	0.95	0.37
2+ prior prison sentences in the past five years vs. 0	1.96	1.64	0.33
1+ prior prison sentences for serious assault vs. 0	0.48	0.11	0.37
1+ prior suspended sentences vs. 0	0.37	0.54	−0.17
1+ prior breaches of violence orders vs. 0	0.32	0.23	0.09
1+ prior suspended sentences for serious assault offences vs. 0	0.22	0.32	−0.10
1+ prior supervised suspended sentences for serious assault vs. 0	−0.11	−0.42	0.31
1+ prior community service orders for serious assault offences vs.0	0.41	−0.47	0.88*
1+ prior good behaviour bonds for serious assault vs. 0	0.17	−0.41	0.58
1+ prior prison sentences for serious assault vs. 0	0.31	−0.04	0.35

Difference between the coefficients is significant at the 5% level.

Table 5 shows the difference in coefficients between the Indigenous and non-Indigenous models. Three variables show a significant interaction effect: whether or not the offence was domestic violence related (Indigenous offenders were more likely than non-Indigenous offenders to receive a prison sentence), whether or not the offence occurred after 2011 (non-Indigenous offenders were more likely to receive a prison sentence after 2011) and whether or not the offender had received a community service order for a serious assault offence in the previous five years (non-Indigenous offenders who received a community service order were more likely to receive a prison sentence than Indigenous offenders). Given this, the marginal probabilities for different offenders were recalculated using these two models to show the effect of the significant interaction effects, shown in Figure 2.

Figure 2.

Marginal probabilities of offenders by Indigenous status accounting for interaction effects.

Figure 2 shows that the significant interaction effects have only a small impact on the difference in likelihood of imprisonment. The largest difference found was for offenders committing a domestic violence offence, in which Indigenous offenders were roughly 1.25 percentage points more likely to receive a prison sentence than non-Indigenous offenders (compared to the base-case for offenders, in which Indigenous offenders were less than one percentage point more likely). The interaction effects for year-group of offence and the presence of previous community service orders for serious assault offences both increased the likelihood of non-Indigenous offenders receiving prison sentences relative to Indigenous offenders. When the offence occurred after 2012, the difference in the likelihood of imprisonment was negligible across Indigenous status, whereas the presence of a previous community service order for a serious assault offence actually made a non-Indigenous offender more likely to be imprisoned than an Indigenous offender.

Discussion

Given the substantial difference in incarceration rates between Indigenous and non-Indigenous offenders in NSW, it is vital to look at factors which could be influencing this, including sentencing bias. This is especially important given the inconsistency in previous research between Snowball and Weatherburn (2006, 2007), on the one hand, and Jeffries and Bond (2015) and Donnelly and Poynton (2015), on the other.

Our binomial GEE regression models showed that Indigenous status of the offender had a significant effect on the probability of receiving a prison sentence over a suspended sentence or supervised bond. This effect persisted as more covariates were added to the model, supporting the findings of Donnelly and Poynton (2015). Furthermore, separate modelling of both Indigenous and non-Indigenous offenders showed an interaction effect between Indigenous status and whether or not the offence was domestic violence related, again supporting Donnelly and Poynton (2015).

Unlike Donnelly and Poynton (2015), however, the effects of Indigenous status in the current study are comparatively small. Once all covariates have been controlled for, the difference in the probability of imprisonment between Indigenous and non-Indigenous assault offender is less than one percentage point (as opposed to approximately 11 percentage points for the demographics only model). This highlights the importance of including comprehensive, sentence relevant controls in any analysis of the effect of race (or Indigenous status) on sentencing. The larger effect of Indigenous status on imprisonment risk found in the Donnelly and Poynton (2015) and Jeffries and Bond (2015) studies would appear to be due to their more limited controls. This said, it must be emphasised that the current study still finds a significant interaction between Indigenous status, offence type and penalty choice. The effect may be small but Indigenous offenders convicted of assault are more likely to receive a prison sentence than non-Indigenous offenders if the assault is domestic. To that extent our results support the interpretation given by Jeffries and Bond (2015, p. 476) of their results: namely that Australian courts in recent times have been treating domestic violence in Indigenous communities as a fundamental cause of Indigenous community dysfunction and, as a result, imposing tougher penalties on Indigenous offenders convicted of domestic violence offences. We agree, moreover, that imprisoning a higher proportion of Indigenous domestic violence offenders is not likely to do much to improve the safety of Indigenous victims of assault. It is bleak consolation to note that our results seem to show that after 2011 courts seem to have become more willing to impose a prison sentence on non-Indigenous offenders.

The interaction of Indigenous status with past receipt of a community service order is interesting and admits of at least two explanations. Snowball and Weatherburn (2007) in their analysis of the effect of racial bias in sentencing found that courts placed less weight on the prior criminal record of the offender in the case of Indigenous offenders than non-Indigenous offenders. Snowball and Weatherburn (2007) suggested that this may reflect the courts’ sensitivity to the economic and social disadvantage experienced by Indigenous Australians (echoing the findings by Jeffries and Bond, 2012, p. 23). It is possible that the current finding (that courts place less weight on the fact that an Indigenous offender has had a community service order than if a non-Indigenous offender has had such an order) reflects a similar tendency. An alternative possibility is that the effect is an artefact of sample size. The number of Indigenous offenders who had previously received a community service order (71) was comparatively small. In the circumstances, it would be unwise to place too much weight on the finding.

Taken as a whole, the current findings give little reason to believe that the gross disparity between Indigenous and non-Indigenous rates of imprisonment is explicable in terms of racial bias in sentencing. The overwhelming weight of evidence suggests that Indigenous/non-Indigenous differences in rates of offending (and re-offending) account for most, if not all of the difference in imprisonment rates (Weatherburn, 2014). Of course, this does not mean that the criminal justice system as a whole is free from racial bias. Police, for example, may be biased in the way they exercise their discretion to refuse or grant bail or in the decision about how to proceed in relation to a suspected offender. These are possibilities that deserve further research. The key question for policy makers is where best to apply their efforts to reduce Indigenous over-representation in prison. Indigenous Australians are known to fare far worse on most of the factors associated with higher rates of offending, including drug and alcohol abuse, child neglect and abuse, poor school performance and unemployment (Weatherburn, 2014). Reducing the influence of these factors would certainly reduce Indigenous over-representation in prison. Whether heightened judicial awareness of the possibility of racial bias in sentencing would have a similar effect is far less clear.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Appendix 1

References

Anthony

(2013) Indigenous people, crime and punishment, London, UK: Routledge.

Anthony

Bartels

Hopkins

(2015) Lessons lost in sentencing: Welding individualised justice to indigenous justice. Melbourne University Law Review 39(1): 47–76.

Australian Bureau of Statistics. (2009). National Offence Index, 2009. Retrieved from http://www.abs.gov.au/ausstats/abs@.nsf/Products/1234.0.55.001∼2009∼Main+Features∼Overview?OpenDocument.

Australian Bureau of Statistics. (2013). Estimates of Aboriginal and Torres Strait Islander Australians, June 2011. Retrieved from http://www.abs.gov.au/ausstats/abs@.nsf/mf/3238.0.55.001.

Baldry

Cunneen

(2014) Imprisoned Indigenous women and the shadow colonial patriarchy. Australian and New Zealand Journal of Criminology 47(2): 276–298.

Blagg, H., Morgan, N., Cunneen, C., & Ferrante, A. (2005). Systemic racism as a factor in the over-representation of Aboriginal people in the criminal justice system (Report to the Equal Opportunity Commission and Aboriginal Justice Forum, Melbourne). Retrieved from Victorian Equal Opportunity and Human Rights Commission website: http://www.humanrightscommission.vic.gov.au/our-resources-and-publications/reports/item/1391-systemic-racism-as-a-factor-in-the-over-representation-of-aboriginal-people-in-the-victorian-criminal-justice-system-sep-2005.

Bond

Jeffries

(2010) Sentencing Indigenous and non-Indigenous women in Western Australia’s higher courts. Psychiatry, Psychology and Law 17(1): 70–78.

Bond

Jeffries

(2011a) Indigeneity and the likelihood of imprisonment in Queensland’s adult and children’s courts. Psychiatry, Psychology and Law 19(2): 169–183. doi:10.1080/13218719.2010.543757.

Bond

Jeffries

(2011b) Indigeneity and the judicial decision to imprison: A study of Western Australia’s higher courts. British Journal of Criminology 51(2): 256–277.

10.

Cunneen, C. (1992, June 23–25). Judicial racism. Paper presented at The Australian Institute of Criminology Conference: Aboriginal Justice Issues, Canberra. Canberra, Australia: The Australia Institute of Criminology.

11.

Cunneen

Allison

Schwartz

(2014) Access to justice for Aboriginal People in the Northern Territory. Australian Journal of Social Issues 49(2): 219–240.

12.

Donnelly, N., & Poynton, S., (2015). Prison penalties for serious domestic and non-domestic assault (Bureau Brief no. 110). Retrieved from http://www.bocsar.nsw.gov.au/Documents/BB/Report-2015-Prison-penalties-for-serious-domestic-and-non-domestic-assault-bb110.pdf.

13.

Jeffries

Bond

(2009) Does Indigeneity matter? Sentencing Indigenous offenders in South Australia’s higher courts. Australian and New Zealand Journal of Criminology 42(1): 47–71.

14.

Jeffries

Bond

(2012) The impact of Indigenous status on adult sentencing: A review of the statistical research literature from the United States, Canada and Australia. Journal of Ethnicity in Crime and Justice 10(3): 223–243.

15.

Jeffries

Bond

(2015) Taking the problem seriously? Sentencing Indigenous and non-Indigenous domestic violence offenders. Australian and New Zealand Journal of Criminology 48(4): 463–482.

16.

Lockwood

Hart

T. C.

Stewart

(2015) First nations peoples and judicial sentencing: Main effects and the impact of contextual variability. British Journal of Criminology 55(4): 769–789.

17.

Miethe

T. D.

Hart

T. C.

Regoeczi

W. C.

(2008) The conjunctive analysis of case configurations: An exploratory method for discrete multivariate analyses of crime data. Journal of Quantitative Criminology 24(2): 227–241.

18.

NSW Bureau of Crime Statistics and Research. (2016). New South Wales Custody Statistics Quarterly Update September 2016. Retrieved from http://www.bocsar.nsw.gov.au/Documents/custody/NSW_Custody_Statistics_Sep2016.pdf.

19.

NSW Sentencing Council. (2009). Sentencing for alcohol-related violence. Retrieved from http://www.sentencingcouncil.justice.nsw.gov.au/Documents/sentencing_for_alcohol-related_violence.pdf.

20.

Paternoster

Brame

Mazerolle

Piquero

(1998) Using the correct statistical test for the equality of regression coefficients. Criminology 36(4): 859–866.

21.

Snowball, L., & Weatherburn, D. (2006). Indigenous over-representation in prison: The role of offender characteristics (Crime and Justice Bulletin No. 99). Retrieved from http://www.bocsar.nsw.gov.au/Documents/CJB/cjb99.pdf.

22.

Snowball

Weatherburn

(2007) Does racial bias in sentencing contribute to Indigenous overrepresentation in prison? Australian and New Journal of Criminology 40(3): 272–290.

23.

Snowball

(2008) Does the lack of alternatives to custody increase the likelihood of a prison sentence? Crime and Justice Bulletin 111, Sydney, Australia: NSW Bureau of Crime Statistics and Research.

24.

Weatherburn

(2014) Arresting incarceration: Pathways out of Indigenous imprisonment, Canberra, Australia: Aboriginal Studies Press.