THE STATE AND
AFRICAN INDEPENDENT CHURCHES IN BOTSWANA PART III Wim van Binsbergen 
homepage  index page
Botswana state and churches  Part I  Part II  Part IV  Part V
2.
Towards a quantitative profile of Botswana churches (b)
Aspects
of registration among the African Independent churches.
When the
preceding analyses are repeated for our 233 African Independent
churches only, the following patterns become discernible:
Membership and year of registration. With
regard to membership and year of registration the relatively poor
quality of the data only allow us to contrast two registration
categories: the registered churches and those which saw their
registration cancelled; within this narrow framework, there turns
out to be a statistically significant association between
registration status and size of church membership: the registered
churches tend to be much larger than the cancelled ones.[1] That the registered African
Independent churches also have a significantly later date of
registration than those which saw their registration cancelled[2] appears to be a different problem: if
we look at registration, and subsequent cancellation of
registration, as a cycle into which an increasing number of
African Independent churches are drawn in their contact with the
Registrar of Societies, it is clear that those churches which
were registered at an earlier also have the greatest chance of
having already reached the later stage of that cycle, i.e.
cancellation.
Registration status and number of congregations: total,
urban and rural. Among the African Independent
churches, there is no longer a statistically significant
association between number of congregations, and registration
status.[3] The relationship between number of
urban congregations and registration status is still
statistically significant, now highlighting particularly the
difference between registered and neverregistered churches, with
the cancelled churches somewhere in between.[4] However, with regard to number of
rural congregations, registration status no longer makes a
difference.[5]
Registration status and urban orientation. Registration
status and ‘urbanity’ of the African Independent
churches is again significantly associated: the registered
churches score highest, followed by the cancelled ones, while the
neverregistered churches close the line.[6] With regard to the
‘rurality’ scale this pattern is of course identical
but reversed.[7]
Factor analysis of the variables in
the data set (for African Independent churches only)
The above
results show the same few variables from a number of
complementary points of view. It is useful to attempt to arrive
at a more comprehensive view. Factor analysis constitutes a
powerful statistical technique that allows us to reduce the
number of variables in a data set and to identify underlying
factors which may in complex ways influence the behaviour of the
surface variables and their interaction. For instance, if
churches would tend to register immediately after emergence, and
if churches would tend to grow at the same rate, then any
association found between date of registration and church size
would have to be attributed to an underlying factor ‘date of
origin’, not directly measured in the data collection.
Factor analysis upon the variables in the data set would
mathematically construct such a factor, calculate the
‘loading’ (between 1 and +1) of each surface variable
upon that still anonymous factor, and allow us to suggest the
nature of by an inspection of the pattern of loadings allow us to
suggest the nature of that fact, e.g. ‘date of origin’.
For students of African divination this technique is easy to
understand, inspiring and exhilarating.
Factor analysis upon the variables in our data set comes up
against the relatively poor quality of the data: the correlation
matrix on which the analysis is based is only meaningful if
missing cases have been deleted ‘listwise’, i.e. if a
missing value on only one of the variables leads to exclusion of
that case from the entire analysis. One strategy is to exclude
variables with many missing cases from the analysis, thus
reducing but not invalidating the matrix. Not only quality of the
data collection but also the specific design of the analysis is
involved here: e.g. churches which were never registered will
inevitably show a missing value on the variable ‘year of
registration’ regardless of the quality of the data. After
some trial and error I limited my analysis to the African
Independent churches in the data set, basing it on 13 variables
and 93 cases which had no missing values on any of these
variables. Most of these variables have been discussed above;
some however particularly refer to Francistown: measuring a
church’s presence (at the congregation and headquarters
level) in that town and in the rural region which surrounds it,
and contrasting that information with other towns and other rural
areas in Botswana. In view of the nationallevel focus in this
paper findings relating to these variables have not been
presented here, but the variables as such are part of the data
set and can help to highlight such underlying factors as it
contains. Finally, either of the pair of variables
‘urbanity’ and ‘rurality’ had to be omitted
in order to avoid problems of multicollinearity: spurious results
due to the confusion of deliberately constructed arithmetical
relations between variables on the one hand, and stochastic
association on the other.
Factor analysis[8] yielded the results presented in table
4.
The following abbreviations have been used for the names of
variables: NOBRANCH, number of branches; URBELS, number of urban
branches in Botswana outside Francistown; URBRA, total number of
urban branches in Botswana; RURBRA, total number of rural
branches in Botswana; RURELS, number of rural branches in
Botswana outside the Francistown region; RURFT, number of rural
branches in the Francistown region; MEMBERS, membership; YEARREG,
year of registration; CHUFT, does this church occur with at least
one urban or rural branch in the Francistown region?; HQFT, does
this church have its headquarters in Francistown?; URBFT, does
this church have an urban branch in Francistown?; URBANITY,
number of urban branches as fraction of total number of branches;
RIG, as stated above, is a slightly modified form of the variable
REG = registration status. With the exception of RIG, all these
variables have been measured on an interval or a dichotomous
nominal scale, which justifies their inclusion in the matrix. RIG
measures a church’s positive interaction with the Registrar
of Societies on an ordinal scale, from ‘exemption’ (1),
via ‘registration’ (2), ‘doubt/has to give proof
of existence’ (3) and ‘registration cancelled’ (4)
to ‘never registered’ (5); strictly speaking such an
ordinal variable does not belong in a factor analysis matrix, but
the intuitive conceptual unilinearity of the scale, and the
variable’s behaviour in the analysis once entered, yet would
appear to justify its inclusion.

ROTATED LOADINGS 


FACTORS 


1 
2 
3 
4 
NOBRANCH 
0.974 
0.146 
0.107 
0.036 
URBELS 
0.936 
0.005 
0.180 
0.029 
URBRA 
0.925 
0.185 
0.206 
0.046 
RURBRA 
0.924 
0.107 
0.309 
0.027 
RURELS 
0.889 
0.007 
0.299 
0.012 
RURFT 
0.736 
0.333 
0.241 
0.052 
MEMBERS 
0.690 
0.142 
0.034 
0.319 
YEARREG 
0.594 
0.004 
0.232 
0.147 
CHUFT 
0.297 
0.923 
0.028 
0.011 
HQFT 
0.237 
0.890 
0.155 
0.053 
URBFT 
0.360 
0.809 
0.192 
0.092 
URBANITY 
0.081 
0.015 
0.931 
0.097 
RIG 
0.044 
0.011 
0.091 
0.961 
VARIANCE EXPLAINED BY ROTATED
COMPONENTS 
5.977 
2.497 
1.322 
1.075 
PERCENT OF TOTAL VARIANCE
EXPLAINED 
45.979 
19.208 
10.166 
8.266 
Table 4. Factor analysis on the data
set: African Independent churches only (high loadings underlined)
The four
rotated factors together explain more than five sixth (83.619%)
of the variance in the data set for African Independent churches,
which is a very high percentage. The nice grouping of the
variables with regard to their loadings on the factors makes it
rather easy to interpret them.
Church size. Factor 1 clearly amounts to an
overall factor church size, as directly
reflected in a church’s number of total, rural and urban
congregations, and its membership. It is noteworthy that also
year of registration should have a significant loading on this
factor: the larger the size of an African Independent church, the
earlier its date of registration. Pending the collection of
further data on churches’ dates of origin and rates of
growth, we might suggest various interpretations for this
association: considering the truly exponential growth of
Independent Churches in Botswana over the past quarter of a
century, older churches have had more time to grow large. And, as
a complementary explanation, larger churches are more
conspicuous, cannot escape the Registrar of Societies’
attention, and also have specific problems (the acquisition of
building plots, conflicts over movable and immovable property in
case of church schisms, the organization of raffles, etc.) for
which registration is required and is positively sought by them;
in the light of the latter explanation it might even be said that
registration is a precondition for African Independent churches
to grow large in the first place. The fact that the loading of
the registration date on factor 1 is relatively slight suggests
that factor 1 primarily measures numbers of people, and not
(contrary to my example when introducing factor analysis) a time
dimension concealed by membership figures which might increase
over the years merely as a function of time.
Francistown bias. Factor 2 seems to reflect
little else than the inevitable Francistown bias
in my data: although the majority of data also for cancelled and
neverregistered churches derive from other parts of the country,
intensive fieldwork in one particular town and region
fortunately cannot fail to yield data of a precision and scope
not available from nationallevel bureaucracies and surveys.
Meanwhile it remains remotely possible that, apart from the data
collection factor, Francistown is genuinely exceptional with
regard to Botswana Independency, as it has been with another type
of Botswana voluntary associations: political parties. In a
country whose Protectorate status largely reduced it to a remote
labour reserve far away from direct confrontation with capitalist
relations of production, Francistown has a unique history of
capitalist encroachment and exploitation (through the early
mines, the massive passage of labour migrants from all over
Southern Africa, and the iron rule of the Tati Company which for
almost a century monopolized and alienated land, agricultural
production and trade in the Francistown region).[9] It is no accident that most political
movements in Botswana originated in this town. It was also (cf.
Lagerwerf 1982; Chirenje 1977) the first place for African
Independent churches to become active, in the beginning of the
twentieth century. However, any specific Francistown factor that
cannot be relegated to the Francistown bias in my data collection
can only be identified once we are satisfied that for other parts
of Botswana data have been collected of the same overall quality
as for Francistown. It is perhaps regrettable that the
Francistown bias in my data seems to be responsible for 20% of
the variance in the data set; on the other hand, we are lucky
that that factor has been made explicit, and that it leaves more
than 80% of the variance to be explained in more systematic
terms.
Rural orientation and its complement, urban
orientation. That it is not the urban nature of such
of the Francistown environment that is involved in Factor 2, is
clear from the minimal loading of the ‘urbanity’
variable on this factor. Factor 3 meanwhile is largely confined
to this variable, and therefore should be taken to measure the
rural orientation of the African Independent churches
in the data set: when ‘urbanity’ is high, Factor 3
(because of its negative sign) is low.
Bureaucratic recognition. The last factor
which was sufficiently powerful to be retained in the analysis is
Factor 4, which again has a significant loading for only one
variable: RIG; when RIG is low (when the church is registered, or
better still exempted), Factor 4 is high, and it can be said to
measure the degree of positive interaction with the Registrar of
Societies’ office, or let us say bureaucratic
recognition.
It is important to realize that, as mathematical constructs,
these four factors have been computed to be be mutually
uncorrelated and irreducible. In other words, while the factor
analysis identifies
(a) church size,
(b) rural orientation
and
(c) bureaucratic
recognition
as major
dimensions along which any African Independent church in Botswana
can be located in our data set and hopefully in social reality
too, such statistical associations between these dimensions as we
may find in the present data set will remain slight: the factor
analysis shows that in principle these factors are independent.
homepage  index page
Botswana state and churches  Part I  Part II  Part IV  Part V
[1]
INDEPENDENT SAMPLES TTEST ON ‘MEMBERS’
GROUPED BY ‘REG’
GROUP 
N 
MEAN 
SD 
1.000 
77 
979.013 
1916.456 
4.000 
33 
251.061 
355.683 
SEPARATE VARIANCES T = 3.207 DF =
87.4 P = .002
POOLED
VARIANCES T = 2.161 DF =
108 P = .033
123 missing cases
[2] INDEPENDENT
SAMPLES TTEST ON ‘YEARREG’ GROUPED
BY ’REG’
GROUP 
N 
MEAN 
SD 
1.000 
144 
80.319 
5.401 
4.000 
43 
76.930 
3.173 
SEPARATE VARIANCES T = 5.129 DF = 119.8
P =0.000
POOLED
VARIANCES T = 3.914 DF =
185 P =0.000
46 missing cases
[3]BARTLETT TEST FOR HOMOGENEITY OF GROUP
VARIANCES, CHISQUARE = 203.155, DF= 2, P = .000

ANALYSIS OF VARIANCE 

SOURCE 
SUM OF SQUARES 
DF 
MEAN SQUARE 
F 
P 
BETWEEN GROUPS 
98.204 
2 
49.102 
1.405 
.248 
WITHIN GROUPS 
7303.267 209 
34.944 



REG 
reg 
canc 
never 
NO CASES 
133 
40 
39 
MEAN 
2.887 
2.100 
1.128 
21 missing cases
[4]BARTLETT TEST FOR HOMOGENEITY OF GROUP
VARIANCES: CHISQUARE = 44.606, DF= 2, P = 0.000

ANALYSIS
OF VARIANCE 

SOURCE 
SUM OF SQUARES 
DF 
MEAN SQUARE 
F 
P 
BETWEEN GROUPS 
20.269 
2 
10.135 
3.455 
.033 
WITHIN GROUPS 
607.259 207 
2.934 



reg 
reg 
canc 
never 
no cases 
132 
40 
38 
mean 
1.477 
1.175 
.658 
23 missing cases
[5]BARTLETT TEST FOR HOMOGENEITY OF GROUP
VARIANCES: CHISQUARE = 176.207, DF= 2, P = .000

ANALYSIS OF VARIANCE 

SOURCE 
SUM OF SQUARES 
DF 
MEAN SQUARE 
F 
P 
BETWEEN GROUPS 
27.333 
2 
13.667 
.640 
.528 
WITHIN GROUPS 
4482.441 
210 
21.345 


REG 
reg 
canc 
never 
NO CASES 
135 
40 
38 
MEAN 
1.393 
.925 
.474 
18 missing cases
[6]BARTLETT TEST FOR HOMOGENEITY OF GROUP
VARIANCES: CHISQUARE = 3.246, DF= 2, P = .197

ANALYSIS
OF VARIANCE 

SOURCE 
SUM OF SQUARES 
DF 
MEAN SQUARE 
F 
P 
BETWEEN GROUPS 
1.185 
2 
0.593 
3.265 
.040 
WITHIN GROUPS 
37.581 
207 
0.182 


reg 
reg 
canc 
never reg 
no cases 
132 
40 
38 




mean 
.73 
.6 
.553 
23 missing cases
[7]BARTLETT TEST FOR HOMOGENEITY OF GROUP
VARIANCES: CHISQUARE = 3.246, DF= 2,
P = .197

ANALYSIS OF VARIANCE 

SOURCE 
SUM OF SQUARES 
DF 
MEAN SQUARE 
F 
P 
BETWEEN GROUPS 
1.185 
2 
0.593 
3.265 
.040 
WITHIN GROUPS 
37.581 
207 
0.182 


reg 
reg 
canc 
never reg 
no cases 
132 
40 
38 
mean 
.27 
.4 
.447 
23 missing cases
[8] The
rotation method used is known as varimax.
The criterion eigenvalue for retention of
factors was set at .9; 1.0 would have been more elegant,
statistically, but would have sacrificed the important fourth
factor.
[9] Cf.
Kerven 1977; Mogotsi 1983; Mupindu 1983; Schapera 1943, 1971;
Tapela 1976, 1982; Werbner 1970, 1971.
homepage  index page
Botswana state and churches  Part I  Part II  Part IV  Part V
page last modified: 20000517 19:55:48