Our empirical approach is affected by reasoning regarding the potential relationship between thriving and surviving activities of households discussed in the introductory section of this paper. In the choice of our method of estimation, we rely on the survey outcomes presented earlier, in which we identify that households reported to use thriving and surviving activities in the crisis period, and used both categories of activities separately as well as in combination. In addition, the modelling framework that we use assumes that both thriving and surviving activities employed by households in the surveyed SEE region can differ according to observed factors (e.g. macroeconomic environment, household economic performance and different dimensions and levels of social capital) as well as non-observed systematic ones (i.e. endogenous), which influence the model. Furthermore, this modelling framework enables us to test some equation-specific (i.e. thriving or surviving specific) determinants, which are identified as being important in the process of estimation. Following a methodological approach used by Efendić, Pugh, and Adnett,^{9} we investigate the correlation between household thriving and surviving activities, their common and specific observed, as well as unobserved, influences. We model these relations by using a system of regression equations, which is estimated as a seemingly unrelated bivarate probit model (SUPM). The main feature—and advantage—of this approach is that we can capture more complex influences in the model than by using a simple simultaneity. There is no model that is perfectly correct, but we believe that our approach seems to be a better choice than a model based on simple mutual causation.^{10}

We will estimate SUPM with the following model specification:
$$\begin{array}{}thrivin{g}_{1}={\hat{\beta}}_{1}+{X}_{K}\cdot {\hat{\beta}}_{1,K}+{\hat{u}}_{1}\end{array}$$(1)
$$\begin{array}{}survivin{g}_{2}={\hat{\beta}}_{2}+{X}_{K}\cdot {\hat{\beta}}_{2,K}+{Z}_{2,B}+{\hat{\beta}}_{2,B}+{\hat{u}}_{2}\end{array}$$(2)
$$\begin{array}{}\hat{\rho}=Cov({\hat{u}}_{1},{\hat{u}}_{2})\end{array}$$(3)

This is a cross-section model with two dependent variables. In Equation (1) *thriving*_{1} denotes an index capturing thriving activities of SEE households, while in Equation (2) *surviving*_{2} codes for household surviving activities. $\begin{array}{}{\hat{\beta}}_{1}\phantom{\rule{thinmathspace}{0ex}}\text{\hspace{0.17em}and\hspace{0.17em}}\phantom{\rule{thinmathspace}{0ex}}{\hat{\beta}}_{2}\end{array}$
are the intercepts in Equations (1) and (2), respectively; $\begin{array}{}{\hat{\beta}}_{1,K},\phantom{\rule{thinmathspace}{0ex}}{\hat{\beta}}_{2,K},\phantom{\rule{thinmathspace}{0ex}}{\hat{\beta}}_{2,B}\end{array}$
are vectors of coefficients to be estimated. Equation specific explanatory variables (capturing specific determinants of household surviving activities) in the second regression are denoted as *Z*_{2B} (1×*B*). *û*_{1} and *û*_{2} are potentially correlated error terms and they include unobserved influences that may contribute to the joint determination of thriving and surviving activities. In Equation (3), the parameter $\hat{\rho}$
can be interpreted as the correlation between the unobservable explanatory variables of the two equations.^{11}

One of the most important statistical checks at the very beginning is to test the statistical significance, as well as the sign and magnitude of the coefficient ϱ. This test has two possible outcomes in this case:

*ρ* = 0 Unobservable influences on household thriving and surviving activities are not associated in the manner suggested by this model; hence, two separate models for investigation of these two activities are needed. This also means that households did not systematically use both types of activities, but that these are mutually exclusive and should be treated separately.

*ρ* ≡ 0 Unobservable influences on household thriving and surviving activities are associated; hence, this model is an appropriate statistical generating mechanism. There is an endogenous link between thriving and surviving activities affected by common, specific and unobserved influences in the model presented above. This finding also implies that thriving and surviving activities of households are not mutually exclusive.

Following good practice, we present descriptive statistics of the variables used in the modelling procedures (including later sensitivity analysis) in in the appendix.

Our dependent variables are constructed as aggregated indices proxying household thriving activities (those reported in ) and household surviving activities (those reported in ). Before creating an aggregated index for these two dependent variables, we conducted a factor analysis that suggested for both cases to combine responses from and into one single factor.^{12} They are combined and for the purpose of this type of model, organised as dummy variables denoting values as reported in the previous table.

The literature review established the importance of household economic performance for involvement in thriving and/or surviving strategies. Accordingly, in both equations, we control for the level of household income reported by the respondents (*incomehi*) as well as the change in the economic situation of households over a period of five years (*ecworse*). It is important to control for both influences, since the current level of income can be fully different in comparison to the situation before the crisis; hence, it is important to control the relative change of their economic situation. We expect that households that have higher levels of income are more likely to be involved in more thriving activities and fewer surviving activities. Correspondingly, those households that reported a negative change in their economic situation are more likely to be involved in surviving activities.

It is well known that the macroeconomic environment is an important determinant of the average household performance, in particular as we have a sample with significant differences between countries in the achieved level of economic development. In that respect, we control for the macroeconomic performance of these countries (*gdppc*) by using the GDP per capita level (divided by 1,000). We argue that this is a good proxy not only for the current level of development, but this indicator captures the entire history of time-varying growth performance.^{13} Moreover, focussing on per capita values means that the relative size of the surveyed SEE countries is taken into account, as well as the possibility that economic data are driven by countries.^{14} We expect that those SEE countries that have higher levels of GDP per capita, hence better macroeconomic performance, are more likely to have households engaged in thriving activities than those countries with a smaller level of macroeconomic output. In relative terms, the level of GDP per capita of these countries did not change over the crisis period.

There is extensive and multidisciplinary literature examining the role of social capital in the everyday life of citizens and households.^{15} It is not that easy to capture this effect as a result of different theoretical approaches, different dimensions of social capital as well as different levels of social capital. However, in our research and based on available data, we capture three dimensions with corresponding levels of social capital. Firstly, we focus on the general role of social capital in the model by controlling for the level of generalised trust (*gentrust*), since trust is seen as a key dimension of social capital in most of the literature.^{16} Our measure captures the trust in unknown individuals as a reflection of confidence in wider social norms, which is the expectation of accepted behaviour of individuals in society in general.^{17} For the purpose of this research we may treat it as a macro-level of social capital.^{18} The second dimension that we use is institutional trust (*instrust*), which is trust in the functioning of the institutional framework including formal rules, organisations and enforcement mechanisms—this in consideration of the definition of formal institutions by Douglass North and the World Bank.^{19} This dimension of social capital may be treated as a meso-level of social capital. Finally, informal institutions, which include unwritten rules, codes, norms of behaviour and networks, are usually a neglected dimension in empirical research, primarily because of the problem of measurement.^{20} We overcome this limitation and include a proxy that captures informal networks based on contacts with different public institutions at the disposal of the respondents (*infcont*). We argue that this is a micro-dimension of informal institutional^{21} social capital at the disposal of the individuals being interviewed, and as such it is important to be included in this investigation. Since more social capital and social interaction is usually seen as an economic advantage of households and individuals, we expect that greater social capital at all examined levels will be associated with more success of households in terms of their greater involvement in thriving and less involvement in surviving activities.

The model that we use enables us to investigate specific determinants linked to the equations in focus. In our case, the statistical test suggests that household surviving activities are systematically affected by four additional determinants; unlike the thriving activities of households (i.e. if we include these independent variables in both equations then they become statistically insignificant, while the model diagnostics become weaker). At a very general level, this can also be treated as an interesting finding—those families that are in a worse economic position and are involved in household surviving activities systematically rely more on additional factors or activities than those in the thriving sample. Simply, this indicates that ‘coping/getting by’ activities are more challenging; they involve more actions and stamina than ‘thriving’ strategies. However, the systematic influences linked to the surviving activities of households include: social interaction within households (i.e. social capital), size and type of households, and additional household productive activities. It is interesting to note that the role of household social capital (*hsocial*) has been identified as important, but solely at the household level (inside the family). In addition, these are also households that report systematic reliance on additional productive activities (*activb*), which are additional activities undertaken next to regular jobs (including e.g. construction, plumbing, wiring, agriculture, etc). These activities were aggregated into a single factor since our factor analysis suggested that they can be combined.

In order to take into account the heterogeneity of the data that may be linked to the countries in focus (i.e. considering their differences in achieved level of economic development), we estimate a robust standard error, in which countries are defined as clusters. This applies to all the estimated models below.

Finally, we estimate our baseline model focussing on the discussed determinants, although individual factors such as gender, age, education and marital status may be important as well. Since this is a household level investigation we do not include these determinants in the baseline specification, but as part of our robustness checks.

The results from the SUPM baseline model estimation are presented in , together with the statistical diagnostics.

Table 5 Results from the baseline SUPM model (cluster-robust inference).

Following good practice of empirical research,^{22} we start our explanation by firstly focussing on the model diagnostics. The Wald test for the joint significance of our independent variables included in the model, rejects the null hypothesis at the highest level of statistical significance, namely that these variables are jointly equal to zero (p=0.000). Next, we rely on the Likelihood-Ratio test to investigate whether the coefficient $\hat{\rho}$ is equal to zero, which is the main test of statistical validity of the estimated model. If this coefficient equals zero, then we cannot rely on this statistical generating mechanism, and instead need to estimate two separate (probit) models for the two dependent variables. The result of the Likelihood-Ratio test implies that the SUPM model is an appropriate estimator for the examined links; hence, thriving and surviving activities of households are not two separate and exogenous concepts, but they are endogenously linked in our model.

The $\hat{\rho}$ coefficient is estimated with the highest level of statistical precision (p=0.000) confirming that we have a proper model. In addition, this coefficient is estimated with a negative correlation coefficient (-0.18) suggesting that more thriving activities of households are systematically associated with fewer surviving activities, and vice versa. It does not mean that households will not rely on both strategies; indeed, they can use both of them and the model identifies this mode of linkage, but also systematic regularity in the model—more thriving and less surviving activities as the general pattern. Later, we examine the models for combinations of households that reported to use only one of these activities, i.e. either thriving or surviving. To obtain the complete picture from this model, and to make interpretation understandable, the next step is to consider the observed joint and specific determinants, which we investigate by estimating the marginal effects of each variable on the probability that households are involved in thriving and surviving activities ().

Table 6 Marginal effects of the SUPM model – thriving and surviving activities.

We find that the majority of independent variables are statistically significant in their relationship with thriving and surviving activities of a household. A qualitative interpretation of the household economic performance, macroeconomic performance and social capital follows.

Household economic performance has the highest magnitude in the model. There is a 16% higher probability that households with higher incomes (*incomehi*) are associated with thriving activities in the period of crisis and less with surviving activities, in comparison to households with lower levels of income. In addition, these households that responded with a negative change in their economic situation over the period of crisis (*ecworse*), are associated less with thriving and more with surviving activities; with a high negative magnitude of 10%. All in all, these results are as expected and they underline the importance of economic performance of households in a period of crisis, which was linked to their activities that followed—either more thriving or surviving.

Macroeconomic performance indicates that the level of economic development (*gdppc*) is estimated with a positive sign, as expected, but it is not precisely measured (statistically not significant). Accordingly, there is no systematic effect of different macroeconomic performance on household activities in the region. Although some SEE countries are more developed than others, household activities during the crisis period are not explained or systematically influenced by this effect. Rather, it is explained with some other joint and specific influences that are similar between countries with different levels of economic development. However, this variable serves its statistical purpose to capture any cross-country economic effect in the model.

Social capital reveals that the different dimensions appear to be important for the examined household activities. After the household economic performance, the highest effect in the model is obtained for generalised trust (*gentrust*, at 7%). The positive and statistically significant coefficient implies that a higher level of generalised trust had a positive effect on household performance during the crisis period—more thriving and less surviving. While generalised trust is precisely measured and estimated, institutional trust (*instrust*) is on the borderline of statistical significance, notably with a positive sign and hence there is a positive effect in the model. However, the magnitude of this coefficient is very low (0.6%) suggesting that although institutional trust is associated with more thriving and less surviving strategies, this effect is almost zero. Finally, informal contact networks (*infcont*) are important, with a magnitude of 2%, and a positive effect in the model. Simply, those households that reported having more informal links in different institutions are also more successful in terms of reporting more thriving and fewer surviving activities. This also signifies the importance of informal networks, and generally informal institutions, in everyday life of households in the region.

Next, we look at specific determinants related to household surviving activities only. These are activities that are not relevant for thriving types of households and include additional productive activities, location, size and social capital of households.

Additional productive activity of households (*activb*) is a statistically significant determinant in the model with a negative sign and has the highest specific magnitude at 6%. This result implies that those households reporting more additional activities are also households that reported fewer surviving activities (e.g. necessity to sell assets such as jewellery, car and land). Accordingly, this is a systematic influence in the model and an important response to the crisis by these households.

Rural versus urban household (*rural*) differentiation has a statistically significant effect in the model, a positive sign and magnitude of 2%. The findings suggest that rural households in comparison to urban ones were more involved in surviving activities. Apparently, the crisis seems to have affected more rural areas in the SEE region when we measure the effect through different forced activities of these households.

Size of households (*hsize*) has emerged as an important determinant in the model as well, having a negative sign and rather small magnitude of 1%. The negative sign implies that bigger households, on average, were less involved in surviving activities than smaller households. Bearing in mind the importance of social capital, informal networks and additional activity in the model, this finding is not surprising.

Household social capital (*hsocial*) is a statistically significant effect of social interactions within the family for surviving activities. The negative and statistically significant coefficient implies that households that reported more social interaction also reported fewer surviving activities, although this effect is very small (0.5%).

Having identified relevant determinants in the model, and especially different dimensions and levels of social capital, we combine these three dimensions (i.e. general trust, institutional trust and informal contact networks) to obtain a visual interpretation of their effects in the model. We estimate this interaction by augmenting the baseline model—as this procedure takes a rich variety of direct and indirect influences of these variables into account. All three variables are now set to binary to make this interaction feasible and to facilitate interpretation. Interestingly, all of the combinations of interactions are statistically significant at the highest level of significance, indicating that there is interaction between the different dimensions of social capital. A visual interpretation of the all three combined factors is presented below (Figure 2).

Figure 2 Interaction of general trust, institutional trust and informal contact networks –Pr(Thriving=1, Surviving=1).

Starting with the upper figure, the grey line identifies for those households with higher generalised trust and more informal links at different institutions. There is approximately a 35% probability that these households will be engaged more in thriving and less in surviving activities. The black line (left side) shows those households with no general trust and no informal contacts with institutions have the smallest probability (around 15%) of combining more thriving and fewer surviving activities. Notably, regardless of what the generalised trust is, this probability does not change much; hence, this conclusion is primarily led by the effect of having informal contacts or not. We will not comment the other combinations as they can be discerned from the figure.

Overall, the previous figure shows that different dimensions of social capital examined through *generalised trust*, *institutional trust* and *informal institutional contacts*, both individually and in combination, moved households in the surveyed SEE region towards greater probability of being in a better position during a period of crisis (more thriving and less surviving activities). However, there is a higher effect of generalised trust and informal institutional contacts than there is of formal institutional trust.

To sum up, by way of our investigation we identify three important findings:

Households in the surveyed SEE region overwhelmingly relied on thriving and surviving activities to overcome the negative consequence of the latest crisis. We identify that these activities were used separately by some households, but also in combination by other households. Thriving and surviving activities were identified to be determined by a number of systematic and endogenous influences, including in particular the household economic performance, social capital performance and household characteristics. Household surviving activities are identified to be under additional specific influences, which implies that these actions are more challenging and thus included more responsive actions.

In terms of observed determinants in the model, our findings suggest that the most important factor affecting whether households were more engaged in thriving or surviving activities is the economic performance of households, primarily in terms of the household income level as well as in terms of the change in the economic situation. This finding implies the importance of economic determinants for different activities used by households.

We also identify the underlying importance of different social capital dimensions for households in the surveyed SEE region and the activities they are involved in. In particular, we identify that generalised trust (macro-level) and informal contact networks existing on the ground (micro-level) are relevant. Simply, the trust in society is a factor positively linked with household performance in terms of employing more thriving and fewer surviving activities. In addition, the existence of informal contact networks at different institutions, which is a type of informal institutional support, was even more important in contributing to better outcomes of households during the period of crisis. Overall, we find that these social capital dimensions, both individually and in combination, move households in SEE regions towards greater probability of being in a better position during a period of crisis (i.e. more thriving and fewer surviving activities).

Although our main interest was to develop a model that takes into account households relying on both thriving and surviving activities, we also developed combinations of models for households using these activities singly (either thriving or surviving), but not in combination ( in the appendix). The results revealed that surviving activities were affected by the same list of determinants discussed earlier, with the following findings: lower income, worse economic situation, lower GDP, fewer informal contacts at institutions, fewer social interactions, smaller households size and fewer additional productive activities are all linked to more surviving activities of households during a period of crisis. Additionally, households in rural locations used surviving activities more. The highest marginal effect was obtained for household income (11%) and additional productive activity (9%). The same conclusion applies for thriving activities with the reverse sign, which is as expected. To sum up, economic performance of households, social capital and household characteristics remain to be important influences for households that rely on thriving or surviving activities only.

As part of our robustness procedure, we estimated a parsimonious model by excluding all variables that are not statistically significant at the conventional 10% level (i.e. excluding *gdppc* and *instrust*). After this change in specification, we did not identify significant differences in magnitudes, signs and statistical significance of the estimated coefficients. Hence, our results are robust to this change in specification. In addition, one may ask why we excluded certain determinants from the thriving equation and link them to the surviving equation only. We also estimated a model in which all determinants were used in both the equations, but the main results still hold while the variables that were included in the thriving equation did not reach a level of statistical significance at the 5% level. We believe that this is due to the specification that does not fit the data well, and thus we interpret the baseline model as more credible. We arrived at the same conclusion when we estimated a model that includes individuals’ characteristics, reported in . Finally, we controlled for the ‘do not know’ responses, but the main results were unchanged in terms of sign and significance, and even magnitude remains very similar.

Following good practice, in the end we may report some concerns regarding our investigation. Firstly, we controlled for different household activities during a period of crisis and distinguished between them based on a theoretical discussion of thriving and surviving household activities. This theoretical distinction may still be challenged and more theoretical underpinnings in the future would be useful. In addition, we combine our specification variables with different levels and different time periods (e.g. pre-crisis influence and postcrisis performance), which would merit more investigation using a dynamic context. In the current model, we cannot control for this. Finally, although we identified non-observed endogenous systematic influences in the model, for policy-making this is ineffectual as results are difficult to interpret.

## Comments (0)