VOTER BEHAVIOUR IN THE METROS: TECHNICAL REPORT

This report accompanies a brief on voter behaviour in the metros in the 2016 local government election.  It details the calculations behind the interpretation contained in the brief.

 

Introduction

Table 1 sets out the allocation of council seats in the eight metros in 2011 and 2016.

 

Table 1 - Metro seat outcomes
2011		ANC	DA	EFF	Others	Total
Tshwane		118	82		10	210
Johannesburg		153	90		17	260
Ekuruhleni		125	62		15	202
Ethekwini		126	43		35	204
Cape Town		73	135		13	221
Nelson Mandela		63	48		9	120
Buffalo City		71	21		8	100
Mangaung		65	25		7	97
Total		794	506		114	1414
Per cent		56,2%	35,8%		8,1%
2016		ANC	DA	EFF	Others	Total
Tshwane		89	93	25	7	214
Johannesburg		121	104	30	15	270
Ekuruhleni		109	77	25	13	224
Ethekwini		126	61	8	20	215
Cape Town		57	154	7	13	231
Nelson Mandela		50	57	6	7	120
Buffalo City		60	24	8	8	100
Mangaung		58	27	9	6	100
Total		670	597	118	89	1474
Per cent		45,5%	40,5%	8,0%	6,0%	100,0%

 

On the face of it, the decline in support for the ANC seems precipitous, but if one adds together the number of seats commanded by the ANC and its three breakaway parties (COPE, EFF and UDM) the total comes to 824 (58.3%) in 2011 and 800 (54.8%) in 2016.  The drop of 3.5% in support for the group contrasts with a gain of 4.7% for the DA and a drop of 1.2% for the other small parties.  The change remains significant, but smaller. 

The purpose of this report is to analyse the change more closely and to bring the results of the 2014 national election into the picture.  The comparison with the national election is facilitated by the fact that the same voting stations were used in the 2014 national and the 2016 local government elections, except in Mangaung.   Some of the voting stations were split into two, or occasionally more, the number of registered voters in each split being separately identified.  This means that geographical areas in 2014 corresponding to the 2016 wards can be identified, since results by voting district are available for both years. 

 

Turnout analysis

For each of the three elections (2011, 2014 and 2016), the percentages of registered voters who voted in each voting district is reported.  The percentages of votes spoilt is also reported.  From these statistics, the net turnout – the percentage of registered voters who cast a valid vote – can be calculated.  It is the net turnout (simply referred as ‘turnout’ in what follows) that forms the basis for analysis in this section.

Table 2 sets out turnout rates for each metro in 2011, 2014 and 2016, separately for wards with an ANC plurality and wards with a DA plurality.  The difference between the ANC and DA ward turnout is calculated.

Not surprisingly, the turnout rate in the 2014 national election is markedly higher than in both the 2011 and 2016 local government elections in all metros.  In all metros except Ethekwini, DA wards saw a higher turnout than ANC wards in all three elections. 

In Tshwane, Johannesburg, Ekuruhleni, Cape Town, Ethekwini and Mangaung, the differences between the DA wards and ANC wards turnouts were higher in the 2014 national election than in both the 2011 and 2016 local government elections.  In Nelson Mandela Bay and Buffalo City, the turnout differences in the 2014 national election were lower than in the 2011 local goverment election.   In all metros except Mangaung, the turnout differential widened in favour of DA wards between the 2011 and 2016 local government elections.  This means that part of the swing towards the DA in the 2016 local government election compared with the 2011 local government election was a consequence of changing turnout differentials.

 

 

Table 2 - Turnout analysis
	ANC wards	DA wards	Difference
Tshwane
2016	51,64%	67,91%	16,27%
2014	70,75%	80,22%	9,47%
2011	50,18%	62,36%	12,18%
Johannesburg
2016	50,79%	63,20%	12,40%
2014	71,56%	76,56%	5,00%
2011	50,95%	59,76%	8,81%
Ekuruhleni
2016	51,60%	65,27%	13,67%
2014	73,68%	77,25%	3,57%
2011	52,70%	60,39%	7,69%
Cape Town
2016	53,73%	66,57%	12,84%
2014	72,34%	75,83%	3,49%
2011	60,44%	65,42%	4,98%
Ethekwini
2016	57,09%	58,89%	1,80%
2014	78,81%	73,11%	-5,70%
2011	58,45%	57,20%	-1,25%
Nelson Mandela
2016	57,08%	68,51%	11,43%
2014	71,62%	76,15%	4,53%
2011	62,05%	65,81%	3,76%
Buffalo City
2016	52,03%	67,32%	15,29%
2014	72,59%	76,85%	4,26%
2011	54,69%	57,58%	2,89%
Mangaung
2016	53,22%	62,70%	9,47%
2014	72,30%	75,67%	3,37%
2011	51,52%	62,13%	10,61%

 

 

Swing analysis

For both the 2104 national election and the 2016 local government election, it is possible to calculate the percentage distribution of votes gained by the ANC, DA, EFF and smaller parties together in each geographical areas corresponding to a 2016 ward.  Can a valid swing analysis be based on these percentages, given the usually substantial different turnouts in general and local government elections?

Two conditions taken together are sufficient for validity:

(i) As a general rule, no-one votes in a local government election without simultaneously being willing to vote in a national election

(ii) As a general rule, the probability of voting in a local government election, given willingness to vote in a national election is independent of party affiliation.

The first condition is likely to be true with few exceptions.  Between a national election and the following local government election, however, the probability of voting in the local government election, given willingness to vote in the national election may well depend on party affiliation.  The question is then this:  Is this dependence a consequence of the violation of condition (ii) or not?  If it is, the comparison will be distorted.  If not, then the differential probability is a result of changing patterns of political behaviour.  In the latter case, suppose the probability dips for supporters of a particular party.  This will indicate a drop in average enthusiasm of that party’s supporters and is validly represented in swing statistics.  It is impossible to tell from available data whether condition (ii) is violated or not and, if its, the extent of the violation.  So the results below should be interpreted with caution. 

What does not matter is the differential turnout across wards as indicated in Table 2, since the analysis is carried out ward by ward.  

There is a further difficulty.  In local government elections, voters have to vote in the voting district in which they are registered.  The rule in national elections is that voters are expected to vote at the voting station where they are registered to vote.  However, voters outside their voting districts on the day of the election may vote at another voting station.  From a swing analysis point of view, this introduces noise into the system, more pronounced at the voting district than at the ward level.

There were changes in pluralities between the ANC and DA in 1.8% of the 2016 ward areas between 2011 and 2014 and 3.6% between 2014 and 2016.  Table 3 sets the changes out by metro.

 

 

Table 3 - Ward plurality shifts, 2014 to 2016
	2011-2014		2014-2016
	ANC to DA	DA to ANC	ANC to DA	DA to ANC	Total wards
Tshwane	1	4	5	1	107
Johannesburg	1	1	6	1	135
Ekuruhleni	1	0	0	0	112
Ethekwini	0	2	9	1	110
Cape Town	0	0	0	0	116
Nelson Mandela	1	0	1	0	60
Buffalo City	0	0	0	1	50
Mangaung	0	2	2	0	50
Total	4	9	23	4	740

 

The swings, in percentages of total votes cast, are set out in Table 4.

 

Table 4 – Distribution of wards by swing between 2014 and 2016

	Swing away			Swing towards
	More  than	4% to 8%	Less than	Less than	4% to 8%	More  than
	8%		4%	4%		8%
Tshwane
2014-2016
ANC	32	47	25	2		1
DA in ANC wards		1		11	45	15
DA in DA wards		1		2	5	27
EFF in ANC wards			14	47	9	2
EFF in DA wards			26	9
2011-2016
ANC	66	13	25	2		1
DA in ANC wards	1	3	5	17	29	14
DA in DA wards	3	1	18	11	4	1
EFF in ANC wards					3	66
EFF in DA wards				22	10	6
Johannesburg
2014-2016
ANC	29	77	25	3		1
DA in ANC wards			2	70	10	7
DA in DA wards	1		6	4	16	19
EFF in ANC wards	1	1	10	61	15	1
EFF in DA wards			23	23
2011-2016
ANC	94	25	12	2		2
DA in ANC wards		1	10	60	13	5
DA in DA wards	4	5	8	24	5
EFF in ANC wards				1	4	84
EFF in DA wards				17	23	6

 

 

 

	Swing away			Swing towards
	More  than	4% to 8%	Less than	Less than	4% to 8%	More  than
	8%		4%	4%		8%
Ekuruhleni
2014-2016
ANC	9	60	36	6	1
DA in ANC wards		1	7	64	5	1
DA in DA wards	1		1	2	23	7
EFF in ANC wards			8	65	5
EFF in DA wards				27	7
2011-2016
ANC	67	20	22	2		1
DA in ANC wards		1	5	63	9
DA in DA wards	1	3	13	9	6	2
EFF in ANC wards				1	9	68
EFF in DA wards				19	8	7
Ethekwini
2014-2016
ANC	42	32	20	8	1	6
DA in ANC wards	1	1	22	40	9	13
DA in DA wards	2	2	2	5	4	7
EFF in ANC wards			16	67		3
EFF in DA wards			7	16
2011-2016
ANC	29	16	23	26	8	7
DA in ANC wards		3	14	47	11	11
DA in DA wards			3	9	2	9
EFF in ANC wards				47	36	3
EFF in DA wards				23

 

 

 

	Swing away			Swing towards
	More  than	4% to 8%	Less than	Less than	4% to 8%	More  than
	8%		4%	4%		8%
Cape Town
2014-2016
ANC	25	33	56	1	1
DA in ANC wards				17	12	6
DA in DA wards	2		13	44	16	6
EFF in ANC wards			6	27	1	1
EFF in DA wards			4	77
2011-2016
ANC	40	26	45	4	1
DA in ANC wards				13	15	7
DA in DA wards	1		60	28	14	8
EFF in ANC wards				2	21	12
EFF in DA wards				76	5
Nelson Mandela
2014-2016
ANC	18	16	22	2	2
DA in ANC wards			2	29	5	2
DA in DA wards		2	1	9	6	4
EFF in ANC wards			6	27	4	1
EFF in DA wards			8	14
2011-2016
ANC	33	7	11	8	1
DA in ANC wards			1	17	19	1
DA in DA wards		3	7	11		1
EFF in ANC wards				4	12	22
EFF in DA wards				21	1

 

 

 

	Swing away			Swing towards
	More  than	4% to 8%	Less than	Less than	4% to 8%	More  than
	8%		4%	4%		8%
Buffalo City
2014-2016
ANC	23	18	8			1
DA in ANC wards			5	29	1	9
DA in DA wards	1				1	4
EFF in ANC wards			3	28	12	1
EFF in DA wards			1	5
2011-2016
ANC	31	7	7	4		1
DA in ANC wards	1	1	9	23	6	4
DA in DA wards	1	1	1	3
EFF in ANC wards				4	10	30
EFF in DA wards				6
Mangaung
2014-2016
ANC	18	10	17	2	1	1
DA in ANC wards	1	1	4	28	2	1
DA in DA wards		1		3	4	4
EFF in ANC wards		1	15	18	3
EFF in DA wards			5	7
2011-2016
ANC	34	2	6	4	1	2
DA in ANC wards	1	2	11	22		1
DA in DA wards	2	5	3	1	1
EFF in ANC wards					3	34
EFF in DA wards				8	3	1

 

 

 

	Swing away			Swing towards
	More  than	4% to 8%	Less than	Less than	4% to 8%	More  than
	8%		4%	4%		8%
All
2014-2016
ANC	196	293	209	24	6	10
DA in ANC wards	2	4	42	288	89	54
DA in DA wards	7	6	23	69	75	78
EFF in ANC wards	1	2	78	340	49	9
EFF in DA wards			74	178	7
2011-2016
ANC	394	116	151	52	11	14
DA in ANC wards	3	11	55	262	102	43
DA in DA wards	12	18	113	96	32	21
EFF in ANC wards				59	98	319
EFF in DA wards				192	50	20

 

The swing away from the ANC is evident everywhere between 2011 and 2016, with the ANC sustaining losses in 90% of 2016 ward areas.  There was a swing towards the DA in wards with an ANC plurality in 2011 in 86% of such ward areas, compared with 51% in ward areas with a DA plurality in 2011.

The swing away from the ANC is also evident between 2014 and 2016, with the ANC sustaining losses in 95% of 2016 ward areas.  There was a swing towards the DA in wards with an ANC plurality in 2014 in 94% of such ward areas, compared with 86% in wards with a DA plurality in 2014.

The EFF improved its position overall, though there was a swing away from it in 152 wards, and in most of the remaining wards it gained between 0% and 4% of votes cast. 

A further analysis is possible.  One can standardise 2011 results by reweighting votes cast in each 2016 ward area in such a way that (i) the 2014 turnout by 2016 ward area is reproduced and (ii) the 2016 turnout by 2016 ward area is reproduced.  The purpose of this standardisation is to remove the effect of differential turnouts on outcomes.  A similar exercise is carried out for the 2014 and 2016 results.  The results are reported in Table 5.

 

 

Table 5 - Standardised turnouts all metros
	ANC	DA	EFF	OTHER
2016 standard
2011	52,74	38,37	0,00	8,88
2014	49,48	37,33	6,62	6,56
2016	43,30	41,71	7,64	7,36
2014 standard
2011	55,52	35,41	0,00	9,07
2014	52,00	34,38	7,03	6,59
2016	45,57	38,80	8,13	7,50
2011 standard
2011	54,02	37,03	0,00	8,94
2014	50,72	36,12	6,62	6,54
2016	44,33	40,50	7,67	7,50
Improved turnout 2016		1,34
Per cent		39%

 

Standardised results are based on the turnouts in each of the three elections.  Thus, the 2016 standardised results indicate what the results in each of three elections would have been if the turnout in each election had followed the 2016 pattern.  Similarly, the 2014 standardised elections indicate what the results in all three elections would have been if all elections had followed the 2014 election turnout pattern.  The 2011 standardised results indicate what the results in all three elections would have been if all elections had followed the 2011 election turnout pattern.

Three features of Table 5 stand out.  The first is that the DA did slightly worse in the 2014 national election than in the 2011 local government election on all three standards.  This contrasts with the gains between the 2014 national election and the 2016 local government election. 

Secondly, the effect of the increased differential in turnout between DA and ANC wards on the increase in the DA vote between 2011 and 2016 can be assessed by comparing the difference between the 2016 DA vote in the 2016 standardisation and the 2016 DA vote in the 2011 standardisation.   This indicates that a swing of 1.34 can be attributed to the increased differential in turnout, 39% of the total DA gain between 2011 and 2016.

Thirdly, while the absolute swing towards the EFF between 2014 and 2016 was smaller than the absolute swing towards the DA in both the 2014 and 2016 standardisations, the swing relative to the 2014 share of the vote was higher than towards the EFF.  

Where does the 2016 local government election imply about ward contestability?  Table 6 presents the number of wards which satisfy the following conditions:

·         ANC wards where the vote for the ANC was less than 60% of the total

·         DA wards where the vote for the ANC was less than 60% of the total

·         ANC wards in which the DA vote was less than 10% of the total

·         ANC wards in which the EFF votes was less than 10% of the total.

Just over a fifth of ANC and DA wards had ANC votes and DA votes respectively that were below 60%.  On the other hand, in 57% of ANC wards, the DA had less than 10% support, and in 42% of ANC wards the EFF had less than 10% support.

 

Table 6 - Contestability of wards, 2016
	ANC	DA	ANC	ANC	ANC	DA
	wards	wards	wards	wards	wards	wards
	ANC<60%	DA<60%	DA<10%	EFF<10%
Tshwane	17	5	18	8	68	39
Johannesburg	22	18	50	15	83	52
Ekuruhleni	17	8	63	20	77	35
Ethekwini	20	13	52	76	79	29
Cape Town	4	1	21	29	81	35
Nelson Mandela	10	2	25	23	37	23
Buffalo City	15	0	28	28	45	5
Mangaung	4	3	32	13	38	12
All	109	50	289	212	508	230
Per cent of ANC wards	21,5%		56,9%	41,7%
Per cent of DA wards		21,7%

 

 

Race, class and voting behaviour

The unit of analysis in this section is the 2016 ward.  We have no data on individuals.

Socio-economic data at the ward level are only available from the 2011 census.  These data are therefore five years out of date.  Yet it is likely expect that the conditions have not changed greatly on average, though there will have been sharper changes in some individual wards, particularly those which have seen a substantial change in the racial composition of the population.

There is also the problem of relating 2016 wards to 2011 wards.  However, one can map 2016 wards on to 2011 wards using voting station numbers.  The criterion for a match is whether a majority of voting stations in 2016 map on to a particular ward in 2011.

Dropping the two wards with IFP pluralities in 2016, 635 (86%) wards out of 740 in 2016 meet this criterion, and it is on this subset that the analysis is based.  In party plurality terms, the subset has roughly the same distribution as the entire set: the ANC has a plurality in 61.6% of the wards in the subset, compared with 62.4% in the entire set.

The race variable is simply the proportion of the ward population age 18 and over that is Black/African.  The class variable is composed from three variables:

·         The employment rate

·         The average years of education among people over the age of 25

·         The logarithm of the median individual income

These variables are strongly correlated.  They have been fed into a factor analysis, and the first factor is interpreted as the class indicator.

Two analyses are then conducted.  The first divides that the race variable into three categories:

·         More than 90% black

·         Between 50% and 90% black

·         Below 50% black,

 

and the class variable (which is measured in terms of z-values, i.e. values from a normal distribution with mean zero and standard deviation one) is also divided into three categories:

 

·         z less than -0.5  (low)

·         z between -0.5 and 0.5  (medium)

·         z above 0.5 (high).

 

These two classifications produce a table with nine cells.  The wards with ANC pluralities are distributed across the cells in Table 7, as are the wards with DA pluralities. 

 

Table 8 considers all votes cast.  It carries out a logit regression[1], with each of the percentage of votes gained by the ANC, DA and EFF as dependent variables, and the race and class variables as independent variables.  It considers three regression models: Model 1 with race only as an independent variable, Model 2 with class only as an independent variable, and Model 3 with both as independent variables.

 

Both tables lead to the conclusion that race continues to outweigh class in determining the voting pattern in wards.  In Table 7, the ANC has pluralities in 97% of the wards that are over 90% black, 68% in wards that are 50% - 90% black and 8% in wards which are less than 50% black.  By contrast, it has pluralities in 91% of the wards that have low socio-economic scores, 69% in wards with medium socio-economic scores and 14% in wards with high socio-economic scores.  Table 7 more strongly indicates the same conclusion.  For each party, the proportion of the variance explained by race is higher than the proportion explained by class, and in the case of the DA, the coefficient of the class variable in Model 2 has the wrong sign.  There is a correlation between the race and the class variables of -0.67, negative because the higher the proportion of black people in the population, the lower tends to be the socio-economic score.  In Model 3, the class coefficients have the expected signs for the ANC and the DA.  The class coefficient in Model 3 for the EFF is positive.  But the addition of the class variable to the race variable adds little to explanatory power (as measured by the increase in variance explained), while the addition of the race variable to the class variable adds considerably more.

 

 

Table 7 – Race and class as determinants of voting behaviour, 2016

Pluralities		Per cent black
		More than 90	50 to 90	Less than 50	Total
ANC
	Low	168	26	2	196
Class	Medium	121	43	7	171
	High	5	11	8	24
	Total	294	80	17	391
DA
	Low	5	0	14	19
Class	Medium	3	18	56	77
	High	0	19	129	148
	Total	8	37	199	244
Per cent ANC
	Low	97%	100%	13%	91%
Class	Medium	98%	70%	11%	69%
	High	100%	37%	6%	14%
	Total	97%	68%	8%	62%

 

 

Table 8 – Race, class and votes, 2016

Regression
Dependent variable
y = ln(anc/(100-anc))
		Model 1	Model 2	Model 3
Coefficients
Race		4,13		3,45
Class			-1,20	-0,37
R-squared		0,85	0,55	0,88
Dependent variable
y = ln(da/(100-da))
		Model 1	Model 2	Model 3
Coefficients
Race		-5,28		-4,29
Class			-0,95	0,53
R-squared		0,86	0,58	0,90
Dependent variable
y = ln(eff/(100-eff))
		Model 1	Model 2	Model 3
Coefficients
Race		2,39		3,17
Class			-0,35	0,42
R-squared		0,55	0,09	0,62
All coefficients significantly
different from zero at
the 5% level

 

 

Conclusions

The conclusions from this study can be summarised as follows:

1. When one considers the ANC alone as a party, the loss of share of seats in all the metros was 10.7% between 2011 and 2016.  When one considers the ANC along with the parties which have split from it since 1994, the loss drops to 3.5% for the group.  The DA has been the beneficiary primarily of both the decrease in share of ANC and splinter parties, and secondarily the decrease in the share of smaller parties other than the ANC and splinters.  The EFF won 8.0% of the seats in all the metros in 2016.

2. In Tshwane, Johannesburg, Ekuruhleni, Cape Town, Ethekwini and Mangaung, the differences between the DA ward and ANC ward turnouts were higher in the 2014 national election than in both the 2011 and 2016 local government elections.  In Nelson Mandela Bay and Buffalo City, the turnout differences in the 2014 national election were lower than in the 2011 local government election.   In all metros except Mangaung, the turnout differential widened in favour of DA wards between the 2011 and 2016 local government elections.  This means that part of the swing towards the DA in the 2016 local government election compared with the 2011 local government election was a consequence of changing turnout differentials.

3. A swing analysis between the 2014 national election and the 2016 local government election and between the 2011 local government election can be carried out on a ward by ward basis.  Its validity depends on the assumption that, as a general rule, the probability of voting in a local government election, given willingness to vote in a national election is independent of party affiliation.  This assumption is impossible to check on the basis of existing information, and swing calculations must be interpreted with caution.  However, carrying out the analysis ward by ward eliminates the effect of differential turnouts within the metros.

4. There were changes in the relative number of votes for the ANC and DA in 13 out of 740 wards between 2011 and 2014, 4 towards the DA and 9 towards the ANC. Between 2014 and 2016 there was a swing of 23 towards the DA and of 4 to the ANC.

5. Between 2011 and 2016, there was a swing away from the ANC in 661 wards.  Within wards with an ANC plurality in 2011, there was a swing towards the DA in 408 wards.  Within wards with a DA plurality in 2011, there was a swing towards the DA in 149 wards.  Between 2014 and 2016, there was a swing towards the EFF in 398 wards with an ANC plurality in 2014 and in 185 wards with a DA plurality in 2014. 

6. Standardisation of results in each election across all metros to 2011, 2014 and 2016 turnouts yields three new insights.  First, compared with the 2011 local government election, the DA lost ground slightly between 2014 national election on all standardisations, while increasing it substantially between 2014 and 2016.  Secondly, nearly 40% of the DA gain between 2011 and 2016 can be attributed to an increase in differential turnouts in wards with ANC and DA pluralities in 2011.  Thirdly, although the absolute percentage swing towards the DA was higher than the swing towards the EFF on both the 2014 and 2016 standardisations, the swing relative to the 2014 vote share was higher for the EFF than for the DA.

7. Four indices of the contestability of wards can be calculated.  Just over one-fifth of the wards with ANC pluralities are held with less than 60% of the vote going to the ANC, and the DA is in virtually the same position.  On the other hand, the DA received less than 10% of the vote in 57% of wards with ANC pluralities.  The corresponding figure for the EFF is 42%, indicating that the EFF achieved more than 10% of the vote in 58% of wards with ANC pluralities.

8. When the composition of wards by race and by class (measured by a factor extracted from three variables: the employment rate among people over 25, the average years of education among people over 25, and the logarithm of median individual income) is considered, it is evident that the racial composition of wards outweighs their class composition.  When both race and class measures are included in a regression analysis, class has a secondary impact on outcomes, in the expected direction in the case of the ANC and DA, and a positive impact in the case of the EFF.  Again these results should be interpreted with caution since the socio-economic information is five years out of date, and it has been possible to match only 57% of 2016 wards with 2011 wards.

9. Looking forward to the national election in 2019 and the local government election to 2021, socio-economic improvement is likely to lead to an improvement in the DA’s performance.  But the effect will be very small, partly because of the secondary role of class in determining electoral outcomes, and partly because expected growth in real per capita income will be very low.  What will matter much more will be (i) the states of the ANC, DA and EFF at the national level in the coming years and (ii) the extent to which there is an improvement or deterioration in service delivery in each metro and which party or parties get the credit or blame.

 

The co-operation of Anele Mtwesi and Kimera Chetty in assembling information from the 2016 local government election is gratefully acknowledged.

 

Charles Simkins
Head of Research
charles@hsf.org.za