Austerity and Anti-Systemic Protest : Bringing Hardships Back In

ed from the variable histories of the countries. We find both confirming and surprising results. First, our expectations regarding the positive impact of short-term hardships on protest are strongly confirmed. Austerity protests are largely explained by the imposition of short-term hardships (p56 = 0.44, all path coefficients are standardized). Additionally, IMF pressure continues to be a clear generator of those short-term hardships (p41 = 0.40), a result that confirms decades of qualitative and quantitative study. Figure 1. Best Multimodel (*p <= 0.05, **p <= 0.01) IMF pressure, however, does not have a direct significant link to austerity protest (p61 = 0.16), which suggests that it is the effects of the policy, not just the often well-publicized policies themselves to which people respond. Additionally, as others have noted (Harvey 2005), neoliberalism has been applied in uneven ways, and the same can be said regarding austerity policies. More surprisingly, neither long-term hardships (p42 = -0.07) nor globalization (p43 = 0.24) have a significant impact on short-term hardships. We will return to this seemingly problematic finding later. Finally, the long-theorized positive effect of civil liberties on protest (Tilly 1974) find no support in the multimodel (p65 = 0.04). As we further unpacked our findings, we found that our multimodel could be refined based on our initial development of the MX model. When analyzing the individual country path models Journal of World-Systems Research | Vol. #21 No. 2 | Austerity and Anti-Systemic Protest jwsr.org | http://dx.doi.org/10.5195/jwsr.2015.15 472 (discussed below), we observed similarities between Argentina and Brazil and between Mexico and Venezuela. To investigate the similarities, we created two additional multimodels (AR + BR and MX + VE). Both of the new multimodels further reinforce our arguments regarding the importance of hardships to protest, but short-term and long-term hardships play roles we did not fully expect. In Figure 2 (a multimodel for Argentina and Brazil), we find additional strong results for our original multimodel, as IMF Pressure again plays an important causal role both directly on protest (p61 = -0.38) and indirectly through short-term hardships (p41 = 0.76 and p56 = 0.70). Short-term hardships play a strong and significant causal role in determining protest events. Notice that the increase in IMF pressure decreased protest (P61 = -0.38) while the associated hardships strongly increased protest (p41*p56 = 0.53). All other effects, except for shocks (E4 and E6) are non-significant. Figure 2. Best Multimodel, Argentina + Brazil (*p <= 0.05, **p <= 0.01) The lack of a significant direct effect from short-term hardship to protest in Mexico and Venezuela country-level models (see below) led us to investigate the direct effect of long-term hardship on protest. In Figure 3 (a multimodel for Mexico and Venezuela), we removed longterm hardships from its exogenous position, and substituted it for short-term hardships. We Journal of World-Systems Research | Vol. # 21 No. 2 | Shefner et al. jwsr.org | http://dx.doi.org/10.5195/jwsr.2015.15 473 continue to find strong and significant direct results for IMF pressure on protest (p61 = 0.80). Neither IMF pressure nor globalization cause long-term hardships (p41 = 0.04 and p43 = 0.33), but long-term hardships do indeed have significant, if somewhat weaker, direct effects on protest (p56 = 0.25). In Mexico and Venezuela also, civil liberties have a strong and significant relation to protest – as civil liberties increased, so too did protest (p65 = 0.58). Figure 3. Best Multimodel, Mexico and Venezuela (*p <= 0.05, **p <= 0.01) Figure 4. Argentina simplified FF Model (*p <= 0.05, **p <= 0.01) Journal of World-Systems Research | Vol. #21 No. 2 | Austerity and Anti-Systemic Protest jwsr.org | http://dx.doi.org/10.5195/jwsr.2015.15 474 To summarize the two disaggregated multimodels: (1) In the AR + BR multimodel, IMF pressure creates short-term hardship which creates protest, and IMF pressure directly creates protest. (2) In the MX + VE multimodel, IMF pressure, long-term hardship and civil liberties each act as exogenous variables creating protest. As we turn to individual country path models, we note that we will elaborate further with two short case studies, but in general the hypothesized relationship of hardships to protest has been confirmed. Argentina (Figure 4) reflects the aggregate multimodel, with some unique differences. Austerity protests respond to short-term hardships (p56 = 0.51), while short-term hardships are in turn explained by IMF pressure (p41 = 0.71). In contrast to the aggregate multimodel, however, protests in Argentina show no direct effect from IMF pressure. Again in contrast, long-term hardships (p42 = 0.49) and our globalization index (p43 = 0.89) demonstrate strong and significant links to short-term hardships as we expected. Civil liberties do not have a significant effect and the coefficient (p65 = -0.44) is in the reverse direction than the literature suggests – here increased civil liberties have a diminishing effect on austerity protest. Figure 5. Brazil simplified FF model (*p <= 0.05, **p <= 0.01) In Brazil (Figure 5), IMF pressure (p61 = -0.84) again has significant direct effects on protest but with a negative sign. IMF pressure does not, however, have significant effects on short-term hardships (p41 = 0.68). Long-term hardships have strong and significant impacts (p42 = 1.07) on short-term hardships, while globalization does not (p43 = -0.85). Finally, short-term Journal of World-Systems Research | Vol. # 21 No. 2 | Shefner et al. jwsr.org | http://dx.doi.org/10.5195/jwsr.2015.15 475 hardships have strong and significant impacts on protest, as hypothesized (p56 = 0.69), and the effect of civil liberties (p65 = -0.41) is not significant. In Brazil, the application of IMF Pressure reduced protest against the background of significant longand short-term hardship. In Mexico (Figure 6), the country in which we developed our model before cross validating in other Latin American countries, we did not find a significant relationship between short-term hardship and protest (p56 = 0.07). This result led us to reconsider the role of long-term hardship in the MX + VE multimodel, that is, substitution along a direct causal path in the model (the modified path model is presented in the Statistical Appendix, see Footnote 3). Figure 6 shows that there are significant links from IMF pressure (p61 = 0.82) and civil liberties (p65 = 0.33) to protests. The link between globalization and long-term hardship (p43 = 0.80) is also significant as are shocks E4 and E6. Figure 6. Mexico Simplified FF model (*p <= 0.05, **p <= 0.01) The same result is replicated in Venezuela (Figure 7). Here, IMF pressure (p61 = 1.14), longterm hardship (p42 = -0.62) and civil liberties (p65 = 1.08) all have significant effects. Globalization, however, does not have significant effect on short-term hardship (p43 = -0.01) and short-term hardship does not have a significant effect on protest (p56 = -0.16). (We have run models for both Mexico and Venezuela where long-term hardship has been substituted for shortterm hardship as in the MX + VE multimodel. Since the results are the same as the multimodel they have not been reproduced here but can be found in the online Statistical Appendix referenced in Footnote 3). Journal of World-Systems Research | Vol. #21 No. 2 | Austerity and Anti-Systemic Protest jwsr.org | http://dx.doi.org/10.5195/jwsr.2015.15 476 The results of the national path models suggest three major conclusions: (1) the same generalized path model predicting a path from hardship (short-term or long-term) to protest fits the four nations examined, but (2) the path to protest in each country is somewhat different, as are (3) the effects of globalization, IMF pressure and civil liberties. Given the different histories of Mexico, Argentina, Brazil and Venezuela during the sample period (1990-2000), there is no a priori reason to believe that one model should fit each country. The fact that one generalized model did fit all the countries is a strong result, and confirms our theorizing of the impact of hardships. In all cases, this result is confirmed. The direct and indirect effects in the model, on the other hand, are rich enough to allow for different paths to protest, and we discuss those more completely in the two case discussions below. Figure 7. Venezuela Simplified FF model (*p <= 0.05, **p <= 0.01)


Introduction
This paper documents index construction and causal modeling for the study of anti-systemic austerity protests in Latin America. First, the theory (Appendix A page 4) and methodology (Appendix B page 7) underlying the study are briefly discussed. Then, an overview of results from the index construction and causal modeling is presented. The detailed results are contained in Appendix F page 23. The conclusions in Section D page 19 suggests that (1) the indexes have adequate (but not perfect) reliability and validity and (2) that the causal models meet all the necessary criteria (normality, lack of autocorrelation and causal ordering) to accept the results of Multimodel testing [9] where models from four high-protest semi-peripheral countries are combined in different ways for overall theoretical tests.

A Theory Explication
For semi-peripheral, open economies in the world-system, a simple model might explain their historical time paths in the late 20 th and early 21 st centuries: hardship and anti-systemic protest were driven by systemic shocks [40] from the world-system, from their domestic economic systems and from domestic political systems. Shocks might include commodity price shocks (oil, natural resources and agricultural products), financial shocks such as the dot-com bubble, debt crises, or the subprime mortgage crisis and political shocks such as regime changes and revolutions. The Shock Theory can be captured with the path diagram in Figure 1 page 4. To the simple Shock model, the directed graph adds a causal mechanism: world-system economic conditions create hardship which results in anti-systemic austerity protest, as suggested by Shefner and Stewart [56]. The basic model in Figure 1 can be elaborated (ignoring the shocks which will be picked up again in the statistical testing) to include intervening effects [48] suggested by world-systems theory. In Figure 2 on page 4 long-term hardship (LTH) and IMF pressure (IMFP) [68] have been modeled as intervening effects between world-system effects and short-term hardship. Also, changes in civil liberties (CIVLIB) and national social investment (NSI) are modeled as intervening effects between short-term hardship (STH) and anti-systemic austerity protest (PROT). The intervening effects might not be operating in all countries and at all times and the directions of the effects (positive or negative) might not always be as expected. In Shefner and Stewart [56], based on studying Mexico during the late 20 th Century, the argument is made that short-term hardship (STH), national social investment (NSI) and changes in civil liberties (CIVLIB) intervene between the effects of IMF pressure (IMFP) and anti-systemic austerity protest (PROT). Increases in national social investment and civil liberties in response to hardship can reduce the need to protest IMF policies. On the other hand, if the IMF austerity measures successfully reduce national social investment and the political system restricts civil liberties, the impetus to protest would be increased. The Shefner-Stewart (SS) model is diagramed in Figure 3 page 5 where ". . . the hardships driven by neoliberal policy. . . energizes democratizing forces that press the state for change"(p. 355) [56].  Also in Shefner-Stewart [56], there is a suggestion that effects are further mediated by the depth and breadth of short-term hardship and by the level of protest severity. These consideration suggest that the short-term hardship (STH) index and the protest index (PROT) might be further broken down into components as displayed in Figure 4 page 5, an idea that will not only be tested here but also helps to clarify the components of the austerity protest index.
One theoretical issue not addressed by the SS model is the direction of causation. In the terminology of General Systems Theory (GST) [64], are the changes in civil liberties and national social investment feedforward or feedback effects? Shefner and Stewart look at increasing democracy in Latin America as happening coincidently with Neoliberal free market policies.
. . . we suggest it is instead the hardships driven by neoliberal policy that energizes democratizing forces to press the state for change. Accordingly, what may appear as a convergence of democratization, neoliberalism and social protest, is actually a causal sequence, beginning with neoliberalism, leading to social protest and ending with democratization. Importantly, as the Mexican transition to political democratization demonstrates, more open politics is not necessarily a cure for longstanding ills of economic inequality [56].
In the Neoliberal model, IMFP, CIVLIB and NSI are exogenous, policy variables manipulated by the state as economic conditions in the world-system (WS) improve after the imposition of austerity. In a political systems (PS) model [22,23,4], changes in CIVLIB and NSI could either be the direct result of hardship (feedforward) or the result of feedback from protest.  [22,23,4] World-systems theory has successfully applied ideas from Hierarchical Systems Theory [25,58,47], specifically that the world-system is coordinated and controlled by the hierarchical relationship between the Core, Semi-peripheral and Peripheral states. There are many other concepts from General Systems Theory [64], Political Systems Theory [22,23,4] and Complex Systems Theory [50,61] that can and will be applied to world-systems analysis in the future. For example, David Easton used the systems model in Figure 5 page 6 to analyze political systems. The feedforward part of Shefner-Stewart could be fit into the input-output framework of Easton's model (the greyed boxes in Figure 5), but since we are not trying to develop a complete theory of the political system (as Easton was trying to do), we do not have to go that far. The feedback part of Easton's model (another form of control), however, will prove useful in developing alternative testable models. Anti-systemic protest is a form of feedback about political outputs. A common misinterpretation of Easton [4] is that because there is a feedback loop in the model, the political system will respond. Easton does not assume that the system has to respond and, as the empirical results below will show, unresponsive political systems can either ignore or repress protests, repression being more likely in peripheral countries ([53] page 37). Anti-systemic protests are still feedback, just feedback ignored. These ideas will be further developed and tested below.
Another theoretical issue not addressed by the static path models and existing theory is the dynamic nature of protest. From watching the unfolding of the European Union (EU) Debt Crisis it would appear that protest occurs in response to specific events, at specific points in the business cycle and during long-waves in world-system development [67]. The Subprime Mortgage Crisis was a cyclical event that hit the US economy in 2007 and spread from there to the EU to include Britain, Ireland, Portugal, Italy, Spain and Greece. In 2012, the Greek government passed austerity measures that resulted in street protests and rioting [39]. These austerity protests were events triggered by world-system shocks during a Kondratieff B-phase, a wave of declining economic performance [67].
The static path diagrams presented above do not make secular or cyclical distinctions 1 but the methodology presented next does make these distinctions. Existing theories of protest have not addressed the dynamic aspects of protest and it proves to be an unavoidable and important issue in the choice of methodology. Although it is possible that there might be secular trends in protests as a result of economic growth and entrenched dysfunctional governments, those cases would seem quite different from shock-driven austerity events and would require different, more complicated dynamic modeling.

B Methodology Details
State-of-the art statistical methodology is based on index construction [30], power analysis [13], generalized cross-validation [28], casual modeling [26,27,48] and testing of model assumptions. Each of these topics is discussed below.

B.1 Index Construction
Theoretical concepts of any generality cannot be operationalized by single variables. There is too much meaning in the concepts to be capture by a single indicator [30]. Multiple indicators help define each of the theoretical constructs presented above. Theoretical concepts developed from multiple indicators, however, cannot be assumed to have inherent validity and reliability. Reliability and validity can be thought of in terms of consistency and accuracy, respectively. If we imagine a target at a rifle range, then reliability is the tightness of the shot pattern and validity is how close the shot pattern is to the center of the bull's eye. For statistical methodologies, there are a number of types of reliability and validity. The ones that are important here involve the percent of variation explained by concept indicators and how well the concept works in statistical models (criterion validity). Concepts that explain more of the variation in the indicators have higher reliability (lower shot dispersion). Concepts that explain more variation in the variables that theory says they should predict have greater criterion validity (closeness to the bull's eye).
The index construction methodology explained below can distinguish between cyclical and secular trends. Each index has at least two components. The components are ordered by percent of variation explained. For time series data the first component, explaining the most variation, is the secular component. The second component, explaining the next largest percent of variation, captures cyclical variation over time (for very cyclical phenomenon such as protest, the first component can also be cyclical). If short-term hardship and protest are cyclical events, then the cyclical components should be used in the final modeling.
By partitioning variation into independent secular, cyclical and error components, the error components can be eliminated from the index for statistical analysis.
The Principle Components [36] measurement matrix (presented below for each index) partitions the true score, T into secular, S, cyclical, C, and error, E, components. Each component is, by construction, independent. The indexes use S and C which are free of measurement error, E. Error partitioning is an added advantage of using indexes over individual indicators. The individual indicators are not free of measurement error. It is important to emphasize that the PCA methodology can be applied to a time series of any length and will still typically produce secular and cyclical components. The theoretical meaning of the secular and cyclical components will not align well with economic theory unless the time period is specified. For example, Kondratieff Waves play an important part of world-systems theory [67] and are thought to be about 50 years in length [41] consisting of two 25-year phases, the A-phase expansion and the B-phase contraction. PCA components from two 25-year samples (if aligned on the start Kondratieff Wave) would see two secular trends, one increasing and one decreasing, respectively, and would not see the complete Kondratieff Wave. PCA components from 11-years of time series data (the present case) will still typically identify secular and cyclical components but will not see Kondratieff Waves. Given the potential for theoretical confusion, it might be better to substitute "oscillatory" for "cyclical" when discussing time series decomposition. However, there is currently no agreed upon usage.
We compute the principal components using the Singular Value Decomposition (SVD). On time series data, the SVD is equivalent to the Spectral Decomposition of X = λ 1 δ 1 δ 1 + λ 2 δ 2 δ 2 + · · · + λ p δ p δ p = U DV [36]. The dominant spectral component has the lowest frequency and explains the most variation while the spectral components of higher frequencies have lower variance. An important result from Hierarchical Systems Theory justifies use of the second principal component: "The middle band of frequencies, which remains after we have eliminated the very high and very low frequencies, will determine the observable dynamics of the system under study..." ([57] page 10).
Four Latin American countries where chosen based on their level of protest. High protest countries selected were Argentina, Brazil, Mexico and Venezuela. Moderate and low protest countries were excluded from this analysis (see Appendix E.2 page 21). In Salvatore Babones' categorization scheme [3], Brazil, Mexico, Jamaica and Panama are in the semi-periphery of the world-system (Argentina and Venezuela are not ranked but are likely also in the semi-periphery). No Latin American countries are among the core countries in the world-system using Babones' methodology. Data were taken from the World Development Indicators published by the World Bank [5] and from other sources (see the codebooks in [55] extracted in Appendix F starting on page 23).
Anti-systemic austerity protest measures were developed by searching over 50 national and international English-language newspapers in the LexisNexis General News database [42] to collect events for the current dataset 2 . Longstanding criteria developed by Walton and Ragin [68] were used to define anti-systemic austerity protest as: "(a) mass actions that (b) specifically addressed austerity policies. . . , and (c) stemming from actions by governments that were strongly urged by international institutions (typically the IMF)" ( [68] page 882). We reviewed thousands of articles and documented events as austerity protests in the nations of interest for the period of 1990-2000. Additionally, cross-national data on each of the countries was collected from a variety of sources including World Bank and United Nations databases, IMF reports, the Polity IV dataset, Freedom House, and the Wall Street Journal.
Each protest event was coded for the number of cities involved, duration, level of violence, form, and participation. Each coded event supplied information about place, participants, form of protest, and precipitating policy imposed by national governments at the urging of international agencies. In the absence of confirming information, we chose not to code the event making the count very conservative. The absolute level of protest, which is certainly underestimated, is not of primary concern in time series analysis. Rather, the relative changes in protest over time are of primary interest. In a simple regression model, for example P ROT = α + βIM F P + E, the intercept term, α, estimates the base level of protest. The intercept is never estimated in the models presented below and is not needed to test the theory being developed. The models estimated here are unaffected by obvious undercounting of protests as long as the undercounting is relatively (but not absolutely) constant over time. If the undercounting is variable, then the constant has an added error term P ROT = (α+ )+βIM F P +E. If the random effects error term is not estimated, then it becomes part of the overall error term P ROT = α + βIM F P + (E + ) which would work against our hypotheses. In other words, if we find significant results even with variable undercounting of austerity protests, the effects are large enough to overcome the added error. The presence of consistent undercounting has no effect on the hypothesis while the effect of variable undercounting works against the hypothesis but does not invalidate it.
From the available data sources, seven indexes were created using Principal Components Analysis (PCA) [36]

B.2 Cross Validation
Theory construction and testing are iterative processes [9]. A theoretical model is based not only on speculation and on other theoretical models, but is also based on contact with data and real historical situations. Evidence is accumulated over time for and against the theory. There is no one single test that either proves or disproves a well-articulated theory. One approach to the iterative process is called generalized cross validation [28]. Using cross-validation, a theory is developed and tested on a small subset of the available data. Independent subsets or samples of the data are then used to test the theory formally. In the present case, Mexico has been used to develop the theory of hardship and anti-systemic protest presented above [56]. In the present paper, the theory is cross-validated using not only Mexico but also Argentina, Brazil and Venezuela, other semi-peripheral countries in the world-system.

B.3 Statistical Power Analysis
Statistical power analysis can be thought as analogous to the power of a microscope. Assume that a laboratory technician is looking for a microscopic disease agent in a Petri dish. If the power of the microscope is too low, the microscopic agent might be missed. If the power of the microscope is too high, the disease agent might also not be visible when looking at individual molecules. In statistical models, power is determined by sample size and effect size. If you are just looking for very large significant effects, you do not need a larger sample. If you are looking for very small significant differences, you will need a much larger sample. If you are trying to test a theory and discriminate between significant and non-significant effects, a large sample size is actually counterproductive. At some large sample size, all effects are significant. Cohen [13] has helped researchers chose appropriate sample sizes by defining small, medium and large effects for typical statistical measures (means, t-statistics, F-statistics, correlation coefficients, etc.). Cohen's approach will be used to understand what types of effects can be observed given the size of the sample (1990-2000, n = 11) available for the present analysis.
The power analysis conducted in Appendix E.4 on page 22 is limited to a univariate correlation coefficient. This means that we will have enough power to test the primary hypothesis which involves the link between STHARD -> PROT. The power analysis allows for the conclusion that, if this single equation is nonsignificant then there is no effect of hardship on anti-systemic austerity protest. Other significant effects in the criterion path model (see Figure 6 page 11 below) can still be accepted as significant findings. What cannot be concluded, given the present sample size, is that non-significant effects in the model are actually zero. Increases in sample size could turn some "marginally significant" results into significant findings (for example, see the IMFP2 -> STHARD2 coefficient in the BR FF Path model in Section H.5 page 9 where p = 0.066, a marginally significant result for an important link). To calculate power against multiple comparisons, a multivariate criterion measure would have to be developed. Morrison [44] presents a number of criterion measures that might be used when it becomes possible to expand the current sample size.
For Multimodel comparisons, the model with the best (lowest) Akaike Information Criterion [1,9,10] is chosen. The AIC selects the simplest model among competitors, that is, the best model (smallest residuals) with the fewest path coefficients. For the development of theory, Multimodel inference suggests that models be evaluated not only by accumulated statistical tests but also by the model's richness and simplicity. The best model to explain anti-systemic austerity protest might be the simple single equation STHARD -> PROT or even E -> PROT (the Shock Model), but the single equation does not have much explanatory richness. The more interesting feedforward equation system WS -> STHARD -> PROT provides a richer understanding of anti-systemic protest. Including the feedback equation system PROT -> STHARD -> WS in the Multimodel comparison brings in potential insights from Political Systems Theory and further advances our understanding of protest. Including additional intervening variables, as suggested by Shefner-Stewart, would be accepted as long as the added model complexity brings added predictive power (a smaller AIC, that is, smaller residuals corrected for the increase in model parameters).

B.4 Testing Assumptions
Statistical tests are based on assuming a theoretical shape for the population probability distribution. Typically, the Central Limit Theorem is invoked to apply normal theory as the population distribution. The Central Limit Theorem [7] proves that adding together indicators, regardless of their basic distributions, will result in a measure which is normally distributed with the classical bell-shaped distribution. The other assumption is independence, typically created by random sampling. In models based on time series data, the independence assumption is often violated by observed autocorrelation, that is correlations over time between model residuals. All these assumptions have to be tested before concluding that the normal distribution can be used for hypothesis testing.

B.5 Causal Modeling
From the SS (Shefner-Stewart) and the PS (Political System) models, six causal models [26,27,48] were developed: a feedforward (FF) model, a feedback model (FB) and a simplified FF model with two variations for each model. The ability to articulate multiple models allows for Multimodel inference (MMI) [9,10]. In MMI, the simplest model with the most explanatory power is chosen as the "best" model among a number of competitors. MMI argues that the purpose of science is not only to test hypotheses but also to articulate and develop models. Models endure and may even be resistant to "definitive" testing because they contain the explanation for what is happening. Statistical tests provide evidence for and against the model but results must accumulate over time. In the present case, all four semi-peripheral Latin American countries have the highest level of anti-systemic austerity protest. The multiple models contain alternative explanations for why. Some model might work better in one Latin American country than another. Some other model might work better in core rather than semi-peripheral or peripheral countries. Over time, the evidence accumulates for one model over another. Results from no single country during one single time period can provide a definitive test of the best model.
The feedforward (FF) model is displayed in Figure 6 page 11. In the FF model, globalization, long-term hardship from world-system position and IMF pressure (from a world-system institution) create short-term hardship. IMF Pressure can also directly create protest. Civil liberties and national social investment are intervening effects between short-term hardship and protest. As will become clear from the discussion of index construction below, the second version of the FF model eliminates national social investment which is shown in dashed lines in Figure 6. The feedback (FB) model is displayed in Figure 7 page 11. In the FB model, protest has an immediate effect on short-term hardship and IMF Pressure with NSI and CIVLIB acting as intervening effects. Short-term hardship also has a direct effect on long-term hardship and globalization. The signs of many of the coefficients are expected to be negative or at least result in negative total effects (following all paths from PROT to the dependent variables [31,38]) to account for negative feedback. In the PS model, protest should result in policy changes that reduce the reasons for the protests. In the simplified FF model ( Figure 8 page 11), national social investment and civil liberties are modeled as exogenous policy variables that can be increased once austerity policies have had their desired effect. Path diagrams model instantaneous causality. Any changes in hardship, for example, should result in immediate increases in the protests. In the real world where data is reported on a yearly basis and effective protests require some time for participants to be mobilized, there will be lags and delays throughout the system. A typical time series model would look at the effect of last year's PROT, for example, on this year's protests in addition to the effect of last year's STHARD. The model might be P ROT t = α + β 1 P ROT t−1 + β 2 ST HARD t−1 + t−1 . A first-order difference equation of this kind is said to test Granger Causality [29]. Lags beyond one period could also be added to such a model. Protests might be more likely in an unresponsive political system that allows grievances to accumulate over many years requiring multiple time lags to model protest. The causal reality for anti-systemic austerity protests probably lies somewhere between the two extremes of instantaneous causes and multi-year, accumulated grievances. The current understanding of causality [48] holds that causes must exist simultaneously or in the prior period to qualify as a valid causal explanation. Causality-at-a-distance is seen as a short-cut explanation that leaves the mechanism by which effects are transmitted unspecified. Events in the 1970's and 1980's may well have caused protest mobilization in the 1990s, but some societal-level mechanism would be needed and that mechanism would require complicated modeling that exceeds our current theoretical understanding. In the modeling below, simultaneous causality and Granger causality will both be tested. Longer lag structures and more complex dynamic causal models will be reserved for future speculations and investigations.

C Results of Estimation
Results are presented for sample selection, index construction, indexes by country, path models by country and an aggregate Multimodel for all the high-protest semi-peripheral Latin American countries. Although the results for the index construction and casual modeling vary across countries, some generalizations can be made taking each index separately across countries.

C.1 Analysis of the Sample
Results for the sample are presented in Appendix E page 21. The data set ran from 1990-2000 (n = 11). Protest totals for all Latin American countries are displayed in Figure 12 page 21. There are peaks in the beginning and the middle of the sample period. The protest totals for each Latin American country in the overall sample are given in Appendix E.2 page 21. Brazil, Argentina, Venezuela and Mexico, in that order, have the highest protest totals. Each is a semi-peripheral country in the world-system. A time plot of protest in the high protest semi-peripheral countries is displayed in Figure 13 page 22. Protest in these countries shows a strong peak in 1995.
The power of a statistical test is a joint function of the sample size and the effect size. By choosing a significance level (α = 0.05 as the probability of a Type I error, that is, incorrect rejection of a true hypothesis) and desired power (β = 0.80 the probability of avoiding Type II error, that is, incorrect failure to reject a false null hypothesis) a sample size and effect size can be computed [11]. The results presented in Appendix E.4 page 22 show that large correlations (r > 0.76) can be identified with an n = 11 (the country-level sample size) and that moderate correlations (r > 0.41) can be identified in the high-protest Multimodel (n = 44). The sample size has adequate power to test the STHARD -> PROT link. There is not adequate power to conclude that other non-significant effects might be zero (see Appendix E.4 page 22 for a more complete explanation).

C.2 Index Analysis
The percent of variation explained by each index is a measure of reliability. The indicators that load (have large coefficients) on each index are typically used to interpret the index and the secular or cyclical time path of the index determines how well it might explain either hardship or protest.
For the GLOB index, the items were always weighted to show that increases in the index over time reflected increasing globalization as defined by the KOF index.

Protest
The first protest index, PROT1, is in all countries an approximately equal weighting of all the indicators. In all cases, the first index explains at least 80% of the variation in the indicators and thus has high reliability. The second, historically unique index, PROT2, is typically dominated by one of the indicators, either CITIES (Argentina, Brazil), FORM (Mexico), and DURATION (Venezuela). The secondary indexes typically explain another 5% of the variation in the indicators. SEVERITY did not emerge as a separate component in the index construction.

IMF Pressure
The IMF pressure index includes different policy measures used by the IMF to impose austerity and/or create changes in the affected economies. Over time, the policy measures tended not to be used together. For example, the use of the extended fund and the use of IMF credit tended to be separate policy instruments. As a result, no one IMF index explained over 80% of the variation in the indicators. For example, IMF credit dominated the second index in Argentina, but not in other high-protest countries. LNQTA dominated the second index in Brazil; EXFD in Mexico; and DBTEX in Venezuela.

Short-term Hardship
The second short-term hardship index was typically dominated by unemployment and the first index was a relatively equal weighting of the indicators with inflation entering negatively. The first short-term hardship index always explained over 80% of the variation in the indicators and thus had high reliability. A separate STH depth index did not emerge from the analysis of the indicators included.

Long-term Hardship
The second long-term hardship index was dominated by unemployment and family work or children working, except in Venezuela where health expenditure dominated the second index. The first index was a relatively equal weighting of items that were available with some entering negatively. Measures of long-term hardship tended to not be consistently available across countries (zero weightings on an indicator mean that the data was missing).

Civil Liberties
The first civil liberties index did not explain 80% of the variation. The second index tended to be dominated by voting indicators. A priori, a stronger correlation between political development measures and voting might have been expected.

National Social Investment
OBD (Overall Budget Deficit) dominated the second NSI index (Argentina, Brazil, Mexico, and Venezuela). Where data was available, the first index tended to be an equal weighting of the indicators but only tended to explain around 50% of the variation in the indicators. The rationale for including OBD in the index was to capture national social investment funded from debt and to relate this component to IMF pressure to reduced social investment and budget deficits. Our assessment of the NSI index was that it needed further development in a more complex model so it was dropped from the path models.

Economy
The economic index was originally developed to measure environmental sustainability. Impact models [14,24,37,65] formed the basis for the IPCC emission scenarios [45]. Impact models are based on the following identity: where q is productivity, e is energy intensity and c is carbon intensity and N is population, Q is aggregate economic output, E is energy use and C is carbon emissions. Impact modeling finds that all these variables interact to describe economic growth. The lower-case intensive variables describe technologies necessary to create sustainability. Variables for the economic index where chosen based on the findings of Impact models. Impact models provide a better basis for economic growth theory than does the neoclassical economic growth model [2] because, unlike the neoclassical growth model, Impact models are based on the Kaya identity which is true by definition (no similar identity provides the theoretical basis for neoclassical growth theory). In addition to the impact variables, the KOF index of globalization [17,18,20,16,19] and a indicator of economic freedom from the Heritage Foundation (http://www.heritage.org/index/) were also added. The KOF globalization index includes 23 variables that measure the economic, social and political dimensions of globalization. The theoretical definition of globalization used in the KOF index is "...the process of creating networks among actors at multi-continental distances, mediated through a variety of flows including people, information and ideas, capital and goods. [21]" Since the present analysis contains many other social and economic measures, in Appendix F.8 page 26 globalization is redefined operationally in terms of only economic indicators. In the country analysis, (Appendices G, H, I and J) the reliability and validity tests of the KOF index are presented under the ECON index for each country.
The ECON index construction results indicate that the variables are highly correlated in Latin America. The first index is generally an equal weighting of all the indicators except economic freedom (negative in Argentina). The first index always explained over 80% of the variation in the indicators. The second index tends to be weighted most heavily on unique historical trends involving globalization, economic freedom, energy use, emissions, and business cycles, that is, cyclical or historically unique aspects of globalization. It is the historically unique or cyclical aspects of globalization that we hypothesize creates hardships. 3

C.3 Country Analysis
For the country analysis, peaks and troughs in the indexes over time will be identified. These trends need to be considered carefully in terms of causality implied in the Theory Section page 4. In time series estimation algorithms, indexes with secular trend tend to go together while indexes with cycles tend to be related with appropriate lags. The path diagram in Figure 1 page 4 should be kept in mind when reviewing the country results to see if causal patterns emerge that support the model. The definition of the indicator variables for each index is given in Appendix F page 23. Model assumptions were tested with two regression models: (1) P ROT 1 = α+β 1 IM F P 2+β 2 ST HARD2 and ST HARD2 = α + β 1 IM F P 2 + β 2 LT HARD2 + β 3 N SI2 + β 4 CIV LIB2 + β 5 ECON 2. The residuals from these two models were tested both for normality using the Shapiro-Wilk test [54] and for autocorrelation (independence) using the Breusch-Pagan test [8]. Granger causality was tested using the general equation Q t = α+β 1 Q t−1 +β 2 X t−1 where Q is one time series variable and X is one other variable in the data set. Only significant results are reported. Finally, each of the six path models is tested and compared using the Akiake Information Criteria (AIC) [1,10]. The best model is chosen as the one with the smallest AIC meaning that the model had the smallest residuals with the fewest variables, in other words, the simplest model with the best predictive power.

Argentina (Appendix G page 27)
The measurement models and time plots for the AR Indexes can be found in Appendix G page 27. The PROT index shows a very cyclical pattern with no apparent trend over time. The IMFP1 index shows secular change over time while the IMFP2 index peaks in 1995. The STHARD1 index shows growth with an asymptote around 1995. The LTHARD1 index shows secular growth over time while the LTHARD2 index peaks in 1996. The CIVLIB1 index shows acceleration after 1998 while the CIVLIB2 index (voting behavior) shows secular trend. The NSI1 index shows secular trend while the NSI2 index shows a trough in 1996. The ECON1 index shows secular trend while the ECON2 index shows a flat peak (a period of high globalization) between 1991 and 1998. In general the cyclical nature of the PROT index and the secular nature of the STHARD index do not suggest that the two are causally related. The correlation between the KOF index of globalization and the GLOB1 component was r = 0.92 suggesting high convergent validity.
Neither the AR Protest model nor the AR Hardship model show either autocorrelation or departures from normality. Granger causality testing identified six significant effects, three on hardship, two on National Social Investment and one on IMFP. The best AR path model was the simplified FF model (AIC = 19.12) in which all path coefficients except from IMFP2 -> PROT1 and from CIVLIB2 -> PROT1 were significant as were shocks to STHARD1 (E6) and PROT1 (E4). In the Feedback (FB) model there were significant negative effects from PROT1 -> STHARD2.

Brazil (Appendix H page 35)
The measurement models and time plots for the BR Indexes can be found in Appendix H page 35. PROT1 shows a peak in 1998 and PROT2 shows a peak in 1995. The IMFP1 index shows secular trend while the IMFP2 index shows a trough in 1995 and the IMFP3 index shows a trough a little latter between 1995 and 1998. STHARD1 shows secular trend while STHARD2 shows a trough in 1995. LTHARD1 shows secular trend while LTHARD2 begins declining in 1997. CIVLIB1 shows secular trend while CIVLIB2 peaks in 1993, NSI1 shows secular trend while NSI2 shows cycles with two troughs, one in 1991 and another in 1997. The ECON1 index shows secular trend while the ECON2 index (globalization) peaks between 1991 and 1996. The peak in PROT1 in the late 1990's might be caught by these trends. The correlation between the KOF index of globalization and the GLOB1 component was r = 0.95 suggesting high convergent validity.
Neither the BR Protest model nor the BR Hardship model show either autocorrelation or departures from normality. Granger causality testing identified six significant effects, three on short-term hardship, one on IMF pressure, one on civil liberties and one on globalization (ECON2). The best BR path model was the simplified FF model (AIC = 24.27) with all effects in the model significant at the p < 0.05 level except IMFP2 -> STHARD2, ECON2 -> STHARD2, and CIVLIB2 -> PROT1. All the shocks were significant. In the FF model there were significant negative effects from CIVLIB2 -> STHARD2 and PROT1 -> IMFP2.

Mexico (Appendix I page 43)
PROT1 peaks in 1995 while PROT2 has both peaks and trend acceleration after 1998. The IMFP1 index shows secular decline while IMFP2 shows a peak between 1995 and 1996 while IMFP3 shows a peak in 1994. The STHARD1 index shows a takeoff after 1995 while the STHARD2 index peaks in 1995. The LTHARD1 index shows secular trend while the LTHARD2 index peaks in 1995. The CIVLIB1 index shows secular trend while the CIVLIB2 index shows a peak in 1994 and a trough in 1998. The NSI1 index shows secular decline (this index could be re-weighted to show secular increases since it is dominated negatively by HEP and PED). The NSI2 (OBD) index shows a peak in 1992. The ECON1 index shows secular trend while the ECON2 index peaks in 1995.
The indexes that peak in 1995 (PROT1, STHARD2 and ECON2) suggest a causal relationship. The correlation between the KOF index of globalization and the GLOB1 component was r = 0.94 suggesting high convergent validity.
Neither the MX Protest model nor the MX Hardship model showed either significant autocorrelation or departures from normality. Granger causality testing identified five significant effects, the only one that did not involve the exogenous variables of the MX path models was IMFP2 -> PROT2. The simplified FF model (AIC = 25.50) showed all effects in the model significant except IMFP2 -> STHARD2 and LTHARD2 -> STHARD2. Notice also the IMFP2 -> STHARD2 was not significant but that the Granger causal relationship was significant suggesting that a dynamic model might fit this relationship better. In the FB model there no significant negative effects.
Since Mexico was the country in which our model was developed (to be cross-validated on other countries) the failure to find a significant effect from STHARD2 -> PROT1 led to the substitution of LTHARD2 for STHARD2 in the modified FF model (Figure 38 page 51). In the modified model, all effects were significant except IMFP2 -> LTHARD2, and GLOB -> LTHARD2.

Venezuela (Appendix J page 52)
PROT1 shows two peaks, one in 1992 and another in 1996 while PROT2 shows multiple peaks in 1991, 1993, 1996 and possibly 1998 before declining. IMFP1 shows a peak in 1992, IMFP2 shows a trough in 1992 and a mild peak in 1995 while IMFP3 shows peaks in 1993 and 1997-1998. STHARD1 shows a trough in 1992 and secular increase after that. STHARD2 shows a trough in 1993 and a peak in 1996. LTHARD1 shows secular trend while LTHARD2 has a peak in 1992, a trough in 1996 and another peak between 1997-1998. The CIVLIB1 shows secular decline while the CIVLIB2 index (voting) shows a peak between 1996 and 1998. The NSI1 index shows secular trend with a peak in 1996. The NSI2 index shows cyclical behavior with a peak in 1996. The ECON1 index shows secular increase while the ECON2 index (globalization) shows peaks in 1992 and 1997. The correlation between the KOF index of globalization and the GLOB1 component was r = 0.95 suggesting high convergent validity.
Neither the VE Protest model nor the VE Hardship model showed either significant autocorrelation or departures from normality. Granger Causality testing identified seven significant effects. Among the endogenous variables there was one significant effect from PROT1 -> LTHARD2. The the simplified FF model (AIC = 35.01) showed all significant except for IMFP2 -> STHARD2, ECON2 -> STHARD2 and STHARD2 -> PROT1. All shocks to the the endogenous variables PROT1 and STHARD2 were significant. In the FB model there were only significant negative effects from STHARD2 -> LTHARD2. The failure to find a significant link between STHARD2 -> PROT1 let to estimation of a modified FF model where LTHARD2 was substituted for STHARD2 ( Figure 47 page 60). In the modified FF model, all effects are significant except for IMFP2 -> LTHARD2 and GLOB -> LTHARD2.

C.4 Multimodel Estimation
The country analysis presented above shows a great deal of diversity among significant effects within the models. The way effects are transmitted from world-system variables to anti-systemic austerity protest show many differences. The generality of the SS model can be tested through Multimodel inference [10] by averaging the correlation coefficients for the best country models and then conducting an overall test. The results of the MM test are presented in Appendix K on page 61.
The simplified FF path model performs acceptably (AIC = 33.12). All effects in the model are not significant at the p ≤ 0.05 except for IMFP2 -> STHARD2 -> PROT1 and the effect of shocks. The simplified Multimodel path diagram is presented in Figure 9 on page 17 (the path diagram is repeated in the appendix Figure 48  The country-level analysis also suggests that AR and BR might be more similar to each other than to MX and VE. To explore these potential similarities, two further Multimodels were constructed based on 22 observations in each. The major difference between these Multimodels is the role played by Long-term Hardship transmitted from the world-system. The Multimodel for

C.5 Summary of Results
The results can be summarized by model and by the variables used in each model.

Summary of Results by Model
Looking across the models, the following pattern of results emerged: • The simplified Feed Forward (FF) model (Figure 8 page 11) was always the best model, that is, the model with the smallest AIC. For Mexico and Venezuela, the best FF model involved substituting long-term for short-term hardship.
• The Feedback (FB) model was not a strong competitor in any of the semi-peripheral Latin American countries.

Summary of Results by Variable
Looking across the variables used in the path models, the following pattern of results emerged: • IMF Pressure was applied unevenly across the countries during the 1990-2000 period and had uneven effects on protest and hardship.
• Some form of hardship, either long-term or short-term, always led to protest.
• Globalization had uneven effects across the countries.
• Granting of civil liberties had uneven effects on protest across the countries.
• Shocks always had a significant effect on hardship (E4) and protest (E6).

D Methods Discussion
Anti-systemic austerity protest is inherently a cyclical, historically unique phenomenon. Neoliberal ideology views austerity measures as shocks meant to return a dysfunctional economic system to a sustained, exponential growth path. The austerity measures are meant to impose short-term hardship and they do. The resulting anti-systemic protests (if any) are seen by Neoliberals as a short-term nuisance to be endured until the economy is growing again. Political Systems Theory views anti-systemic protest as a strong negative feedback effect. If the political system does not respond by increasing civil liberties or increasing national social investment, the political system will be in danger of losing legitimacy. Any increases in national social investment, however, are discouraged by Neoliberal ideology since it is precisely these investments that are thought to be restricting economic growth. Increases in civil liberties (democracy) are thought to result from increases in economic freedom as an outcome of free-market austerity measures rather than the result of rights won through political struggle. This is a complex dynamic to be explained with a causal model. The mix of policy measures and political feedbacks need not play out in a law-like, preordained fashion. Policy measures may or may not be effective. Protests may or may not be met by a political response. The political response may or may not reduce the adverse effects of austerity. The political response can be an attempt to repress the protests. To increase the historical complexity, the cyclical policy measures are also being imposed on a secular pattern of economic growth and hardship. The Shefner-Stewart (SS) model explains protest through a number of intervening measures to include short-term hardship. SS theory is essentially a feed forward (FF) model that does not make an overt distinction between secular increases in protest and cyclical outbreaks of protest. The distinction between secular and cyclical had to be introduced through the methodology of Principal Components index construction on time series data. There are some secular trends in hardship within the sample but protest indexes are mainly cyclical. It may be that variations in the secular trends explain cyclical outbreaks of protest, but this dynamic would be difficult to model using path analysis. Hardship is another matter. There are secular trends in hardship (both short-and long-term) in each of the semi-peripheral Latin American countries presented in Appendices G, H, I and J starting on page 27. The secular trends could effectively be modeled as a result of secular trends in the world-system in a study that was focused solely on hardship.
The failure of protest severity and short-term hardship depth to emerge as separate indicators suggests either that (1) new indicators are needed for these concepts or (2) a larger sample size will be needed to identify subindexes. To break out protest severity as a separate indicator in the present analysis would be to introduce multi-colinearity and measurement error. The STH index could also be expanded to include measures of subsistence to operationalize the Moral Economy literature [52,63]. There may still be measurement problems with some of the intervening variable indexes and there will be analytical challenges establishing the link between indexes and antisystemic austerity protest.
In spite of some open theoretical and measurement issues and in spite of the clear diversity in country-level responses to world-system shocks and cycles, Multimodel inference showed a clear path from short-term world-system cycles through short-term hardship to anti-system austerity protest, confirming the fundamental insight of the Shefner-Stewart model. The absence of strong feedback effects suggests, not surprisingly, that political systems in Latin America are not very responsive. Whether political systems anywhere are very responsive to anti-systemic protests is an important question for future research. Another important question for future research is the secular growth of hardship in semi-peripheral and peripheral countries. These effects are predicted by world-systems theory and are quite visible in the present sample. The sample needs to be further expanded to include all of the late 20 th century. Expanding the sample of anti-systemic austerity protests will be more difficult because of the care and effort that has to be involved coding protest events from newspaper sources. That project, however, can usefully be continued moving back into the 1970's and 1980's in Latin America and forward to the current anti-systemic austerity protests that result from the 2007-2008 world-wide financial crisis.

E Sample Selection
Anti-systemic austerity protests were coded from the LexisNexis database for 19 Latin American Countries (see Appendix E.2 page 21). Countries were selected for the high protest analysis to maximize variation in protest. Countries were divided into three groups: high, medium and low protest. Data collection covered the period from 1990 -2000. Data were taken from the World Development Indicators published by the World Bank [5] and from other sources (see the codebooks in [55]).

E.3 Country Mnemonics
The following two-digit mnemonics were used for each of the high austerity protest countries.

E.4 Power Analysis
All statistical analyses were based on standardized scores and correlation coefficients. For a sample size with 11 observations (the individual country data from 1990-2000) the approximate power using the arctangh transformation in the R package pwr [11] is: n = 11 r = 0.755788 sig.level = 0.05 power = 0.8 alternative = two.sided The power analysis suggests that, using Cohen's terminology [13], only large effects can be identified in the country-by-country analysis while medium effects can be identified in the Multimodel analysis. To have adequate power to detect moderate effects in each country, the sample size would have to be increased (a future data collection project has already been started). To re-emphasize what was stated in Section B.3 on page 9 above, there is enough power to cover a single univariate statistical test of the primary hypothesis, HARD -> PROT. Other links in the model found to be significant can still be accepted. What cannot be concluded, expect if HARD -> PROT is nonsignificant, is that other effects in the model are zero. Some marginally significant results could become significant with a large sample size (for example, see the IMFP2 -> STHARD2 coefficient in the BR FF Path model in Section H.5 page 9 where p = 0.066, a marginally significant result for an important link). It can only be concluded that no support was found for the insignificant path model coefficients given the current sample size. For Multimodel comparisons, we simply choose the model with the best (lowest) Akaike Information Criterion [1,9,10]. The AIC selects the simplest model among competitors, that is, the best model (smallest residuals) with the fewest path coefficients.

F Index Construction
Six indexes were constructed using Principal Components Analysis (PCA). The indexes were PROT = anti-systemic austerity protest, IMFP=IMF pressure, STHARD = short-term hardship, LTHARD = long-term hardship , CIVLIB = civil liberties, NSI = national social investment and ECON = economic index. The printouts below present the measurement matrix and the fraction of variance explained by each index. The measurement matrix contains the weights assigned to each indicator in the index construction. In general, indexes have been weighted such that the largest weight for each index enters with a positive sign. Assigning a positive sign to the highest weighted indicator allows the index time path the be described as being driven by that indicator. For the globalization index (ECON2 or GLOB), the signs of the weights were chosen such that the weight attached to the KOF index of globalization was always positive. The indexes were constructed from the indicators listed in the next section. Missing data was non-linearly interpolated using spline smoothing available in the base R package [51]. Unlike linear interpolation, non-linear spline smoothing does not bias the change scores from year to year but it does reduce the variation (as does linear interpolation). In cases where non-linear spline interpolation did not produce reasonable results (typically when missing data appear at the beginning of a series), the EM algorithm was used [15]. Where data for an entire indicator is missing, the weights in the measurement matrix are zero.

F.1 Protest Index (PROT)
Anti-systemic austerity protests were coded using the LexisNexis General News database [42]. The IMF Pressure index was based on Walton and Ragin [68].

F.7 Economic Index (ECON)
Economic data were taken from the World Development Indicators [5].

F.8 KOF Index of Globalization
To test the reliability and validity of the KOF index of globalization [21], the following measures were used to describe only the economic dimension of globalization. The economic dimension focuses on foreign trade activity and the level of trade restrictions. All data were taken from the World Development Indicators [5].

K Multimodel Averaging, Estimation and Testing
Multimodel averaging [1,10] involves combining models to create generalized tests of theoretical constructs. Each Multimodel is constructed by averaging the correlation coefficients from different sets of countries. The overall Multimodel averaged correlation coefficients from Argentina + Brazil + Mexico + Venezuela. The next two Multimodels presented below averaged correlation coefficients from Argentina + Brazil and Mexico + Venezuela, respectively. The results of the path modeling for each country (Argentina, Brazil, Mexico and Venezuela) suggest that two Multimodels, one for Argentina + Brazil and another for Mexico + Venezuela should be considered. The key difference between the two path models is the type of Hardships that generated protests. For Argentina + Brazil, STHARD is the primary intervening variable between world-system forces and PROT1. For Mexico + Venezuela, LTHARD is the primary intervening variable between world-system forces and PROT1.

K.1 Overall Multimodel
The correlation coefficients from the feed-forward (FF) models for Argentina, Brazil, Mexico and Venezuela were averaged to form a Multimodel correlation matrix. The simple FF model was then re-estimated using the Multimodel correlation matrix. The results suggest that many of the effects that may or may not have been significant at the country level were significant when averaged together. Particularly the effect from STHARD2 -> PROT1 was significant, confirming the Shefner-Stewart model. The transmission paths from world-system total effects 4 in the FF Multimodel are presented in Figure 48 page 62. The paths from IMF pressure through hardship to protest are significant and in the right direction. Effects from hardship and protest shocks are also significant.