Study design
To enable causal inference about the average treatment effect of the Finnish basic income experiment on the use of dental care services, the original experimental design of the experiment is exploited by comparing the study outcomes of the marginally randomized basic income group and control group. Because persons were randomly allocated to the basic income group and to the control group, the groups are similar both in their observed and unobserved characteristics. Thus, any differences in the use of dental care services are attributable to the policy intervention. Being in the basic income system versus being in the existing tax-benefit system is the only characteristic that differs systematically between the study groups.
Outcome variables
In the main analysis, the individual-level outcome variables were defined as (1) having one or more visits to dental care services, (2) the total number of visits to dental care services, and (3) the out-of-pocket expenditure on dental care services during the experimentation period (2017–2018). The large range of outcome variables was chosen in order to capture and differentiate between changes in access (yes/no), volume (number of visits), and consumption patterns (out-of-pocket spending). In addition, the study outcomes were separately defined for public (including primary care and hospital care) and private dental care services to take into account the institutional specificities of the Finnish dental care system.
A visit to dental care may contain several procedures. For this study, a visit was further defined as a single date with oral-health-related procedures, i.e., each person was set to have a maximum of one visit per day and per service provider (primary care, hospital care, private care). Concerning primary care services, the data were limited to contacts that were actual visits by using information on the contact type and excluded, for example, remote contacts by phone or email.
Data on dental care expenses were available only for the private service use. Out-of-pocket expenditure was defined as costs of dental procedures after National Health Insurance reimbursements. The expenses were operationalized as out-of-pocket costs in order to directly measure how much money the participants spend on private care services. In the main analysis, total private dental care expenditure is also reported to evaluate the impact on the National Health Insurance system.
For the outcome variables of the main analysis, a two-year period (2017–2018) was chosen for two reasons. First, it is typical for dentists to recommend booking a visit for regular checkups for every other year only. Second, in 2016, the National Health Insurance scheme was changed to cover visits in the private sector only every other calendar year unless the patient’s health status, verified by a dentist, requires otherwise.
For descriptive purposes, all outcome variables were calculated also for the two-year period preceding the experiment (2015–2016). For further analyses (see
Supplement), annual and monthly outcome variables were calculated for the whole data period of 2015–2019. In addition, outcomes during the experiment (2017–2018) were calculated separately for selected dental procedure categories and for different service providers (private care, primary care, hospital care) to gain information on different types of services, such as preventive, specialized and emergency care.
Visits to private dental care services are not recorded in the data if a person does not apply for National Health Insurance reimbursements. However, the number of unreimbursed cases is likely very small because reimbursements are usually handled automatically when paying for the service. In addition, the use of dental care services provided by the employers is not measured in the data. On the other hand, dental care is rarely included in the health care service contracts of the employers, and thus the employer-provided visits are expected to be very few in the study population.
Statistical approach
The study utilizes the following statististical approach: First, basic descriptives of the background variables of the basic income group and control group are reported in order to describe the study population and to evaluate the success of the randomization procedure in balancing the study groups in relation to characteristics that may predict the later use of dental care services. Second, the average treatment effects on different outcome variables are analysed by estimating the following Ordinary Least Squares (OLS) regression model:
$$ {Y}_{i}=\alpha +{X}_{i}^{{\prime }}\beta +\delta {Tr}_{i}+{\epsilon }_{i}$$
In the equation, \( {Y}_{i}\) represents the examined dental care usage outcomes (one or more visits, number of visits, out-of-pocket costs) measured for the two-year experimentation period 2017–2018. \( {Tr}_{i}\) is the treatment status indicator (basic income groups vs. control group), \( {X}_{i}\) includes the control variables for different background characteristics observed before the experiment, and \( {\epsilon }_{i}\) is the individual-level error term in the model.
The average treatment effect is estimated both with a simple (1) and a multiple (2) OLS regression model with heteroskedasticity-robust standard errors. In a marginally randomized experiment, a simple regression model (1) suffices for estimating the average treatment effect. However, adding baseline variables with predictive power as covariates in the model may increase the statistical power of the estimation [
44]. Covariates in the multiple model (2) include previous unemployment benefit type, gender, age group, having children, having a partner, native language, urbanization level of the place of residence, and previous use of public and private dental care services.
In the multiple model (2), age is categorized into three brackets: 25-34, 35-44, and 45-59. Gender is included as a binary covariate, and native language is coded as official domestic language (Finnish or Swedish) or foreign language. Family structure is measured with the number of dependent children, recoded into a binary variable having or not having children, and with marital status and with information on cohabitation, together recoded into a binary variable having a partner (= married or cohabiting) or not having a partner. Data on the place of residence (municipality) is categorized to urban, semi-urban, and rural municipalities according to Statistics Finland’s classification for year 2016 [
45]. For age, children, partner, and place of residence, information from the end of 2016 is used.
The estimation is complemented with an analysis of effect heterogeneity withing selected subgroups. The analysis is conducted by estimating the simple model (1) separately for each of the subgroups. In addition, regression analyses are conducted separately for visits with surgical and non-surgical dental care procedures and for visits with different types of non-surgical procedures (e.g., examinations, preventive procedures, and restorative treatments) (see
Supplement).
Study population descriptives
The target population of the Finnish basic income experiment composed of low-income individuals. Accoding to the main evaluation study, persons in the study population had 24 days in employment and 286 days in unemployment during year 2016, on average [
42]. Average earnings from employment were 1,900 euros in 2016.
Table
1 describes the socioeconomic and demographic background characteristics of the study groups. The mean annual taxable income, including earnings and taxable social benefits, was around 10,800 euros in the basic income group and in the control group in 2016. The gender ratio was quite equal between the study groups, 48% of the persons being women. The mean age in the basic income group was 40.8 years and 40.4 years in the control group, and 25% of the persons in the groups had other than Finnish or Swedish (official domestic languages) registered as their native language. There are no statistically significant differences between the study groups regarding the listed background characteristics at 5% significance level indicating a successful randomization procedure in balancing the study groups.
Table 1
Background characteristics before experiment (2016)
Annual taxable income, € | 10744.43 | 10826.21 | -81.78 | 0.413 |
Women, % | 47.8 | 47.5 | 0.3 | 0.812 |
Age | | | | 0.294 |
25–34, % | 33.5 | 35.1 | -1.6 | |
35–44, % | 27.4 | 27.1 | 0.3 | |
45–59, % | 39.0 | 37.7 | 1.3 | |
Foreign language, % | 24.6 | 25.4 | -0.8 | 0.398 |
Place of residence | | | | 0.731 |
Urban, % | 79.0 | 79.5 | -0.5 | |
Semi-urban, % | 10.9 | 11.0 | -0.1 | |
Rural, % | 10.0 | 9.5 | 0.5 | |
N | 2000 | 173,222 | | |
In order to further describe the study population, Table
2 summarizes the pre-experimental use of dental care services in the study groups (a two-year period 2015–2016). In both groups, 63% of the persons had a visit in dental care services. About 56% visited primary care, while 6% in the basic income group and 5% in the control group had visits in hospital care. Approximately 11% in both groups used private dental care services.
Table 2
Use of dental care before the experiment (2015–2016) by study group
Visiting | | | | |
Any care, % | 62.5 | 62.5 | -0.1 | 0.957 |
Primary care, % | 55.5 | 55.9 | -0.4 | 0.730 |
Hospital care, % | 6.1 | 5.3 | 0.8 | 0.121 |
Private care, % | 11.1 | 10.9 | 0.2 | 0.781 |
Number of visits to | | | | |
Any care | 5.86 | 5.84 | -0.02 | 0.927 |
Primary care | 5.03 | 5.06 | -0.03 | 0.857 |
Hospital care | 0.14 | 0.16 | -0.02 | 0.368 |
Private care | 0.70 | 0.63 | 0.07 | 0.302 |
Out-of-pocket expenditure | | | | |
Private care, € | 44.49 | 40.87 | 3.62 | 0.445 |
N | 2000 | 173,222 | | |
Number of visits to dental care services during the two-year period was 5.9 in the basic income group and 5.8 in the control group, on average. In the basic income and control groups, the number of visits in primary care was 5.0 and 5.1, in hospital care 0.1 and 0.2, and in private care 0.7 and 0.6, respectively. Persons in the basic income group spent, on average, 44 euros in private dental care services during the two years before the experiment, while the average expenditure in the control group was 41 euros. The differences between the study groups are not statistically significant at 5% significance level indicating balanced study groups needed for a design-based causal inference.
Results: effects on the use of dental care services
Table
3 reports the estimated average treatment effects on the selected outcome variables measuring the use of dental care, i.e., the probability of visiting, total number of visits, and the out-of-pocket expenditure on dental care services. The estimation period covers the whole two years of the experiment (2017–2018), and the effect estimates are provided separately for using any care, public care (including primary care and hospital care), or private care. Table
3 reports the estimates from a simple OLS regression with only the treatment status indicator as a predictor.
Table 3
Average effects on the use of dental care (2017–2018)
Any care | | | | |
P of visiting | 0.634 | -0.016 | 0.011 | 0.132 |
Visits | 5.34 | -0.08 | 0.17 | 0.627 |
Public care | | | | |
P of visiting | 0.570 | -0.027 | 0.011 | 0.017 |
Visits | 4.76 | -0.19 | 0.16 | 0.212 |
Private care | | | | |
P of visiting | 0.114 | 0.013 | 0.007 | 0.072 |
Visits | 0.57 | 0.12 | 0.06 | 0.066 |
Total expenditure (€) | 47.56 | 13.83 | 6.43 | 0.032 |
Out-of-pocket expenditure (€) | 40.58 | 12.09 | 5.65 | 0.032 |
Based on the estimation, we do not find statistically significant effects (at 5% level) on the overall use of dental care services, measured both in probability of visiting and total number of visits. However, we do find a statistically significant negative effect of -2.7% points (-4.7% in relative terms) on the probability of visiting public care (p =.017). The estimated effect on the number of visits to public care is also negative, -0.2 (-4.2%), although the estimate is not statistically significant (p =.212).
The estimated effect on the probability of visiting private care is positive, 1.3% points (11.9%), as is the estimated effect on the number of visits to private care, 0.1 (20.4%), but neither of the estimates is statistically significant (
p =.072 and
p =.066, respectively). However, we find a statistically significant positive effect of 12.1 euros (29.8%) on the out-of-pocket expenditure on private care (p.=0.032). The effect estimate on total expenditure (before National Health Insurance reimbursements) indicates a proportional effect on the National Health Insurance expenditure. Adjusting for the selected background characteristic in the regression produces effect estimates and standard errors of similar sizes as found in Table
3 (see Table
S1 in Supplement).
In sum, we do not find a statistically significant effect on the use of dental care services overall. However, the estimation indicates a negative effect on the use of public dental care and a positive effect on the use of private care. Further graphical examinations and statistical estimations provided in Supplement support the findings of the main analysis.