Page 87 - Contributed Paper Session (CPS) - Volume 6

CPS1829 Lana Clara Chikhungu et al.
needs to justify whereabouts, humiliated, threatened, insulted). Cluster
analysis is performed using the cluster function in Stata 13.0 for Windows
(Statacorp, Tx, USA). Clustering is performed using hierarchical cluster analysis
based on Ward’s distance. To decide on the number of clusters, we use a
combination of indices which indicate optimal clustering pattern ( Calinski–
Harabasz pseudo-F index (Calinski & Harabasz, 1974) and the Duda–Hart
index (Duda & Hart, 1973) and dendritic analysis.
Regression analysis
Once the final cluster profile was selected, membership of a particular
cluster was then used as the dependent variable in a multinomial logistic
regression. A multinomial logistic regression model is appropriate because
instead of running four separate regression models, separate logistic
regression model for each indicator variable are estimated simultaneously.
The model takes the form of equation 1

ln ( ) = ′
=0
= 0 …

In equation 1 the probability of belonging to cluster is denoted as
where there are clusters. The probability of belonging to cluster is
modelled as a logit function of a combination of a vector of coefficients and
associated predictor variables in the vector . − 1 logits are estimated and
the baseline cluster = 0 is omitted to identify the model. (Anderson &
Rutkowski, 2008).

3. Results
Cluster analysis
Five distinct clusters of abuse were extracted based on a combination of fit
statistics and dendritic analysis.
The first cluster is termed no abuse (NA), and contains women who
reported no experience of abuse on any of the response variables. This cluster
comprises 25.39% of the sample.
The second cluster extracted is characterised by Controlling Behaviour (CB).
All women within this cluster report that they have experienced their spouse
demanding knowledge of their whereabouts, but no other form of abuse is
present. This cluster comprises 541 women, or 11.79% of the sample.
The third cluster comprises general controlling behaviour (GCB). Women in
this cluster experience high rates of jealousy from their husbands (84%) and
their husbands demand to know whereabouts (77%), as well as lower level of
accusations of being unfaithful (30%), isolation from friends (22%) and family
(17%). All indicators of physical abuse are low (below 5%). This is the second
largest cluster in the sample comprising 1245 women (27.13%).

76 | I S I W S C 2 0 1 9