The results are presented in three different subsections where the first presents the study population (descriptive statistics), the second present the results from PCA and FA methods, while the third presents the findings of the regression analyses.
Characteristics of the study population
Table 1 and Table 2 present some of the background characteristics of our respondents. Table 1 presents the mean and standard deviations of the background variables, while Table 2 shows how our outcome variable (membership status) differs across different explanatory variables. Our study consisted of 722 respondents, 304 (42.1%) of them being men while 418 (57.9%) were female. The mean age of the respondents was 44.7 years (SD. 13.67). Most of the respondents i.e. 72% had completed primary school education and almost three-quarter were engaged in small-scale farming. The mean household size was 5.4 members (SD. 2.3). Thirty-seven percent of the respondents had a monthly income below 50,000 Tanzanian shillings (TZS), which is equivalent to 22 USD, while 1% had a monthly income above 1 million TZS (435 USD). It also follows from Table 2 that 30% of the respondents reported that their households were enrolled in the iCHF as members, of which 61.5% were female and 39% were men.
Table 1: Characteristics of the study participants
Variables
|
Mean
|
SD
|
Age (years)
|
44.67
|
13.67
|
Household size
|
5.39
|
2.28
|
Monthly income ( in TZS)
|
124358
|
188538
|
Gender (1=female)
|
0.42
|
0.49
|
Marital status (1=married)
|
2.67
|
1.37
|
Religion (1=Christian)
|
0.86
|
0.35
|
Occupation (1=farmer)
|
0.74
|
0.44
|
Education level
|
|
|
No formal Education(1=yes)
|
0.18
|
0.38
|
Primary Education (1=yes)
|
0.72
|
0.45
|
Source: Authors’ calculation based on primary data
Note: Primary data collected from two rural districts of Dodoma (Bahi and Chamwino)
Table 2: Characteristics of the respondents by membership status
Characteristics
|
Member(s)(%)
|
Non-Member(s)(%)
|
Total
|
Age (years)
|
|
|
|
60+
|
39(17.9)
|
61(12.10)
|
100 (13.9)
|
40-59
|
103(47.2)
|
238(47.2)
|
341(47.2)
|
26-39
|
63(28.9)
|
176(34.9)
|
239(33.1)
|
18-25
|
13(5.9)
|
29(5.8)
|
42(5.8)
|
Gender
|
|
|
|
Female
|
134(61.5)
|
284(56.4)
|
418(57.9)
|
Male
|
84(38.5)
|
220(43.7)
|
304(42.1)
|
Education
|
|
|
|
Secondary and higher education
|
28(12.8)
|
47(9.3)
|
75(10.4)
|
Primary education
|
154(70.6)
|
366(72.6)
|
520(72)
|
No education
|
36(16.5)
|
91(18.1)
|
127(17.6)
|
Marital status
|
|
|
|
Unmarried
|
55(25.2)
|
143(28.4)
|
198(27.4)
|
Married
|
163(74.8)
|
361(71.6)
|
524(72.6)
|
Household size
|
|
|
|
≥10
|
10(4.6)
|
20(4.0))
|
30(4.2)
|
7-9
|
56(25.7)
|
122(24.2)
|
178(24.7)
|
4-6
|
112(51.4)
|
261(51.8)
|
373(51.7)
|
≤3
|
40(18.4)
|
101(20.0)
|
141(19.5)
|
Occupation
|
|
|
|
Non-farmer
|
58(26.6)
|
129(25.6)
|
173(25.9)
|
Farmer
|
160(73.4)
|
375(74.4)
|
535(74.1)
|
|
|
|
|
Source: Authors’ calculation based on primary data
Note: Primary data collected from two rural districts of Dodoma (Bahi and Chamwino)
Principal Component and Factor Analysis
We start by reporting the various statistical tests performed before PCA and FA. Results for Bartlett’s test of sphericity, Kaiser-Meyer-Olkin measure (KMO), and Cronbach’s alpha are reported in Table 3. According to the literature[28,29], such diagnostic procedures indicate to what extent PCA and FA are appropriate. We observed that the standard requirements for KMO and Cronbachs alpha (see the right column of Table 3) were fulfilled.
Table 3: KMO measure, Cronbach’s alpha and Bartlett's test of sphericity
S/N
|
Test
|
values
|
requirements
|
1
|
Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy
|
0.815
|
KMO>0.5
|
2
|
Cronbach’s alpha measure of scale reliability
|
0.801
|
α >0.7
|
3
|
Bartlett’s test of sphericity
Chi-square
Degrees of freedom
Significance
|
4892.747
703
p <0.000
|
p <0.05
|
Source: Author’s illustration
Both PCA and FA apply eigenvalues higher than one as the inclusion criteria [30]. According to Costello and Osborne, variables whose loadings are ≥ |0.3| should be retained[25], We also carried out Orthogonal rotation (varimax) to improve the interpretation of the extracted factors.
Our findings on PCA are presented in Table 4. For this method, 10 factors met the eigenvalue criteria and they accounted for 60% of the explained variation. Three of the 10 factors did not fulfill the factor-loading criteria (two or more statements within each factor and a factor loading ≥ |0.3|), leaving us with seven factors that in sum contained 28 of the 33 statements. The number of statements belonging to each factor varied from two to six. The seven factors are quite homogenous in the sense that they include statements that are concerned with similar subjects. The exception is the two statements that are concerned with affordability (price-income considerations) that are grouped into Preferences (S11) and Knowledge (S24). We also observe that the 9 statements that measure the degree of understanding are grouped into three different factors denoted as Understanding, Knowledge, and Awareness.[1] It follows that the most important factor is provider-related (Quality) since accounting for almost 11% of the explained variance. This factor includes statements that all measure various quality dimensions of health care services. The least important factors are the five scheme-related factors of which Convenience is the most important one (7 % of the explained variance). Preferences are the second most important factor since explaining more than 9% of the variance. This factor reflects general preferences as well as alternative strategies to insurance (borrowing and saving) and curing (traditional medicine).
Table 4: Principal Component Analysis (PCA): Household perceptions towards iCHF
S/N
|
Factors and statements
|
The explained variance (%)
|
Factor Loadings
|
P1
|
Quality (health care services)
|
10.6
|
|
S1
|
Healthcare services
|
|
0.76
|
S2
|
Healthcare personnel
|
|
0.72
|
S3
|
Long waiting time
|
|
-0.71
|
S4
|
Reasonable treatment time
|
|
0.71
|
S5
|
Discrimination of members
|
|
-0.65
|
S6
|
Availability of drugs
|
|
0. 56
|
P2
|
Preferences (beliefs and priorities)
|
9.5
|
|
S7
|
iCHF is a loss of money
|
|
0.69
|
S8
|
I save money in case of illness
|
|
0.68
|
S9
|
I borrow money in case of illness
|
|
0.66
|
S10
|
Prefer traditional healers
|
|
0.59
|
S11
|
Low benefit-premium ratio
|
|
0.50
|
S12
|
Insurance brings bad luck
|
|
0.42
|
P3
|
Convenience (iCHF accessibility)
|
7.2
|
|
S13
|
Office hours
|
|
0.83
|
S14
|
Opening location
|
|
0.78
|
S15
|
Card collection
|
|
0.72
|
P4
|
Understanding (iCHF)
|
5.1
|
|
S16
|
Only relevant for chronic diseases
|
|
0.80
|
S17
|
Health is in the hands of God
|
|
0.74
|
S18
|
iCHF is for government workers
|
|
0.39
|
P5
|
Recommendation (iCHF)
|
5.1
|
|
S19
|
iCHF representatives
|
|
0.85
|
S20
|
Relatives and friends
|
|
0.83
|
P6
|
Knowledge (iCHF)
|
4.8
|
|
S21
|
Awareness about the iCHF premium
|
|
0.78
|
S22
|
The iCHF benefits are clear to me
|
|
0.51
|
S23
|
Knowledge about the iCHF scheme
|
|
-0.39
|
S24
|
The iCHF Premium is affordable
|
|
0.38
|
P7
|
Awareness (iCHF)
|
4.7
|
|
S25
|
iCHF is for irregular incomes earners
|
|
-0.65
|
S26
|
I know people that are members of iCHF
|
|
0.57
|
S27
|
Current needs are prioritized
|
|
0.44
|
S28
|
iCHF is like paying taxes
|
|
0.42
|
Source: Authors’ calculation of PCA based on primary data
Note: Primary data collected from two rural districts of Dodoma (Bahi and Chamwino)
The findings for the factor analysis (FA) are presented in Table 5. For this method, four factors were identified that accounted for 91% of the explained variation. All four factors fulfilled the factor-loading criteria and in sum, the 4 factors include 22 of the 33 statements. The number of statements belonging to each factor varied from two to eight. The most significant changes, compared with PCA, are that Preferences (P2) and Understanding (P4) now are collapsed into one single factor denoted as Preferences/Understanding (F2). Furthermore, we observe that; (i) an additional provider quality dimension (facilities, S29) becomes part of Quality (F1), (ii) the affordability statements (S11 and S24) are now ignored, and, (iii) two of the three factors that measured the degree of understanding (Knowledge and Awareness) are now excluded.
Table 5: Factor Analysis (FA): Household Perceptions towards iCHF
S/N
|
Factors and included statements
|
The explained variance (%)
|
Factor Loadings
|
F1
|
Quality (health care services)
|
34.1
|
|
S1
|
Healthcare services
|
|
0.74
|
S2
|
Healthcare personnel
|
|
0.71
|
S3
|
Long waiting time
|
|
-0.63
|
S4
|
Reasonable treatment time
|
|
0.60
|
S5
|
Discrimination of members
|
|
-0.55
|
S6
|
Availability of drugs
|
|
0.67
|
S29
|
Facilities (equipment)
|
|
0.33
|
F2
|
Preferences/Understanding
|
27.4
|
|
S7
|
iCHF is a loss of money
|
|
0.50
|
S8
|
I save money in case of illness
|
|
0.50
|
S9
|
I borrow money in case of illness
|
|
0.60
|
s10
|
Prefer traditional healers
|
|
0.60
|
S12
|
Insurance brings bad luck
|
|
0.55
|
S16
|
Only relevant for chronic diseases
|
|
0.33
|
S17
|
Health is in the hands of God
|
|
0.43
|
S18
|
iCHF is for government workers
|
|
0.54
|
F3
|
Convenience (iCHF accessibility)
|
19.6
|
|
S13
|
Office hours
|
|
0.69
|
S14
|
Opening location
|
|
0.66
|
S15
|
Card collection
|
|
0.53
|
S21
|
Awareness about the iCHF premium
|
|
0.31
|
S30
|
iCHF is a prepayment scheme
|
|
0.36
|
F4
|
Recommendation (iCHF)
|
9.9
|
|
S19
|
iCHF representatives
|
|
0.59
|
S20
|
Relatives & friends
|
|
0.59
|
Source: Authors’ calculation of FA based on primary data
Note: Primary data collected from two rural districts of Dodoma (Bahi and Chamwino)
The three most important factors for FA are Quality (F1), Preferences/Understanding (F2), and Convenience (F3), and they account for about 34%, 27%, and about 20%, respectively, of the total variance. This means that the four most important factors identified for PCA (P1-P4) are also the most important ones for FA, however, for the latter two of the four factors are integrated into one single factor (Preferences/Understanding).
The various perception factors, together with household characteristics, are introduced as independent variables in multivariate regressions where iCHF membership status is the dependent variable. Based upon the statements belonging to each of the factors, we expect positive associations between membership and Quality (P1 and F1), Convenience (P3 and F3) Knowledge (P6), and Recommendation (P5 and F4) while we expect negative associations for Preferences (P2), Understanding (P4) and Preferences/Understanding (F2). As concerning the household characteristics, education, income, and household size are expected to increase the probability of being enrolled in the iCHF.
Regression analysis
The logistic regression results are presented in Table 6. A total of fifteen variables influencing the household membership status were included in the first model and 12 variables in the second model. The first model included seven perception factors identified from PCA combined with eight household characteristics while the second model had 4 perception factors identified by FA and 8 household variables. From Table 6 we observe that 6 out of the 7 perception factors given PCA were significant (Awareness was non-significant) and 2 out of 8 household characteristics variables were significant. For FA, all 4 perception factors were significant and 2 of the household variables were significant.
Table 6: Multivariate Logistic Regression results of perception factors and household characteristics on membership status.
Variables
|
Model 1: PCA Results
|
Model 2: FA Results
|
|
OR*(SE)
|
OR*(SE)
|
Quality P1, F1
|
1.279 *** (0.101)
|
1.464*** (0.129)
|
Preferences P2, F2
|
0.614*** (0.052)
|
0.577*** (0.063)
|
Convenience P3, F3
|
1.402*** (0.128)
|
1.497*** (0.171)
|
Understanding P4
|
0.830 ** (0.061)
|
|
Recommendation P5, F4
|
0.826*** (0.052)
|
0.843** (0.068)
|
Knowledge P6
|
1.390 *** (0.109)
|
|
Awareness P7
|
1.075 (0.078)
|
|
Household characteristics
|
|
|
Gender
|
|
|
Female
|
1
|
1
|
Male
|
0.753 (0.146)
|
0.753 (0.145)
|
Age (years)
|
|
|
60+
|
1
|
1
|
40-59
|
0.571** (0.156)
|
0.567** (0.154)
|
26-39
|
0.459*** (0.136)
|
0.466*** (0.136)
|
18-25
|
0.582 (0.268)
|
0.562 (0.252)
|
Education
|
|
|
Secondary and higher education
|
1
|
1
|
Primary education
|
1.029 (0.325)
|
0.918 (0.282)
|
No education
|
1.268 (0.489)
|
1.049 (0.394)
|
Marital status
|
|
|
Unmarried
|
1
|
1
|
Married
|
1.165 (0.257)
|
1.193 (0.263)
|
Family size
|
|
|
≥10
|
1
|
1
|
7-9
|
0.760 (0.361)
|
0.751 (0.357)
|
4-6
|
0.736 (0.338)
|
0.737 (0.336)
|
≤3
|
0.677 (0.325)
|
0.679 (0.327)
|
Religion
|
|
|
Muslim
|
1
|
1
|
Christian
|
1.119 (0.289)
|
1.162 (0.296)
|
Occupation
|
|
|
Non-farmers
|
1
|
1
|
Farmers
|
0.951 (0.202)
|
0.968 (0.206)
|
Income (in TZS)
|
|
|
1.000.000 and higher
|
1
|
1
|
500.000-999.999
|
0.683 (0.562)
|
0.599 (0.488)
|
100.000-499.999
|
0.480 (0.349)
|
0.416 (0.299)
|
50.000-99.999
|
0.357 (0.264)
|
0.317 (0.231)
|
0 - 49.999
|
0.267* (0.198)
|
0.218** (0.159)
|
Number of observations
|
722
|
722
|
Log-likelihood
|
-391.5037
|
-396.7734
|
Likelihood ratio test
|
84.02
|
77.42
|
Prob >chi2
|
0.000
|
0.000
|
Pseudo R2
|
0.1145
|
0.1028
|
Source: Authors’ calculation of logistic regression based on primary data
Notes: (1) Primary data collected from two rural districts of Dodoma (Bahi and Chamwino)
(2) Significance level: ***(p ≤0.01); **(p ≤ 0.05); *(p ≤ 0.1)
The signs of the factors are as expected except for Recommendation (P5 and F4). The factors that appear to be most important, evaluated by significance levels and the size of the odds-ratios, are Preferences, Convenience, Knowledge, and Quality for PCA while for FA they are Convenience, Preferences/Understanding, and Quality.
Three factors for PCA and two factors for FA have a positive association with enrolment status. For PCA, the odds of a household being enrolled into iCHF, increase by 28%, 40%, and 39% as Quality, Convenience, and Knowledge, respectively, become higher. For FA, the odds of enrolling in the iCHF scheme increase by 46% (Quality) and 49% (Convenience). Factors that are decreasing the odds of enrolling (both for PCA and FA) are; Preferences, Understanding, and Recommendation.
We also observed that two of the eight variables (age and income) are statistically significant in both model 1 and model 2. The odds of being an iCHF member are 51%, 58%, and 44% lower for households whose respondent was aged between 18-25 years, 26-39 years, and 40-49 years relatively to households whose respondent is aged 60 years or older. Regarding household’s income, the odds of being insured by iCHF are 76% lower for households with income between 0-49,999 Tshs, relatively to households with income of 1,000,000 TZS or higher. Contrary to our expectations, household size and education level turned out insignificant.
[1] Both Understanding, Knowledge and Awareness are dominated by statements concerned with measuring the respondents’ understanding of the iCHF scheme, and to what degree they are informed about the contract terms.