Does Adjustment Lending Work? Policy Reforms in the Wake of Program Lending. Abstract

Transcription

1 Does Adjustment Lending Work? Policy Reforms in the Wake of Program Lending Joshua Loud and Daniel L. Nielson Brigham Young University 6 April, 2007 Abstract This paper examines the effects of structural adjustment lending on governments policy choices. Prior research has focused largely on the effects of IMF stand-by agreements on economic growth generally. But the literature has largely neglected a prior step in the causal chain: the effects of adjustment lending on policy choice. And previous scholarship has also paid less attention to the role played by other international financial institutions, including the World Bank, and even more neglected are the roles potentially played by the regional development banks and by bilateral financing. We address both of these issues by using comprehensive data from the Project-Level Aid Database (PLAID), which includes data on adjustment financing from multilateral banks as well as from bilateral sources. We also employ multiple measures of policy reform, including inflation, exchange rate stability, and size of budget deficit. We find that structural adjustment projects do not typically translate into policy reform.

2 Introduction Contradictory stories of structural adjustment pervade the popular press. On the one hand, structural adjustment is feckless, where, as journalist Sebastian Mallaby writes, recipient states have traditionally understood that adjustment loans turn as much on geopolitical concerns and the international financial institutions (IFIs) need to push money out the door to justify their existence as it does on countries objective need and subsequent conformity to loan conditions. The result: Mutually assured hypocrisy... (Mallaby 2004, 46). This story holds that adjustment lending fails especially because most adjustment loans are fully disbursed regardless of whether or not the conditions are met. On the other hand, some critics have claimed that the International Monetary Fund and World Bank have great power to alter government decisions. The problem: they force feed bad medicine that makes sick economies sicker. As Nobel laureate Joseph Stiglitz, the former chief economist at the Bank, writes of the Fund s efforts following the Asian Financial Crisis of the late-1990s, All the IMF did was make East Asia's recessions deeper, longer, and harder (Stiglitz 2000). Defenders of structural adjustment claim that loan conditions are necessary to ensure the long-term economic well-being and fiscal responsibility of recipient governments. Without the Bretton Woods institutions and their management of the international economy, the world economy would be a much less stable place. Unfortunately, careful and extensive social science has done little to resolve this public debate. Some scholars have argued that aid boosts economic growth only when accompanied by the good policies demanded by conditionality (Burnside and Dollar

3 2001, Noorbakhsh and Paloni 2001). Others have held that aid spurs growth regardless of the policy environment (Hansen and Tarp 2001). Still others present compelling evidence that adjustment lending has a negative effect on economic growth, even apart from countries prior circumstances (Przeworski and Vreeland 2001). This prior work has asked a critical question by focusing on the ultimate effects of adjustment lending. But as the unsettled debate shows, agreement among scholars has proven elusive. However, in connecting adjustment programs directly to economic growth, the literature has largely talked past a prior question: Does adjustment lending actually induce policy change? In this paper we examine the effects of structural adjustment lending on governments policy choices. Moreover, earlier studies have usually focused on one or a few international institutions, particularly the International Monetary Fund. However, countries in need of adjustment lending often receive loans from multiple institutions, including the IMF, the World Bank, regional development banks, and other governments bilaterally and the conditional effects of the loans may be magnified if they come from more than one source. We employ econometric analysis using the new Project-Level Aid Database, which provides nearly the full universe of multilateral and bilateral development aid, including adjustment loans, from 1970 to Use of PLAID data allows us to weight each adjustment episode by the number and identity of adjustment lenders involved, and it also allows us to control for the size of each adjustment loan. Prior studies have often examined countries receipt of structural adjustment loans or not, coded as dummy variables thus ignoring variance in terms of size of loans and the degree to which they

4 are coordinated with program lending from other multilaterals and governments. Our PLAID dataset thus allows us to generate measures of adjustment lending that are more complete. Finally, prior work has explored the effects of structural adjustment loans on economic growth or other similarly distant dependent variables. But uncovering factors that affect GDP growth often proves elusive, and thus changes in GDP may not be proximate enough for the effects of adjustment lending to be observed. Instead, we examine the effects of structural adjustment loans on the more proximate dependent variable of policy choices, including stabilization, budget deficits, trade liberalization, and current account deficits. Thus, these more proximate dependent variables, along with more complete data and more nuanced data for each lender, ought to more directly reflect the effects of adjustment lending. To test the core hypothesis that adjustment lending affects policies, we perform pooled time series cross-section analysis on a dataset including 80 countries from 1980 to Here, we test the effects of adjustment loans on overall levels of inflation, budget shortfalls, and exchange rates. The Argument Despite advanced methods and generally shared epistemology, the literature on the effectiveness of foreign aid, including adjustment lending, has reached quite contradictory conclusions. On the one hand, some scholars have found that aid promotes economic growth regardless of the degree of recipient countries conformity to conditions imposed by lenders (Morrissey 2004, Hansen and Tarp 2001). Others have argued that the effectiveness of aid is largely conditioned by governments providing good policy

5 environments, where policymakers heed the international financial institutions (IFIs) demands for market liberalization, stabilization, and sound fiscal management (Burnside and Dollar 2000). Still others find that IMF adjustment programs are related to lower rates of economic growth even when adjusting for the selection bias inherent in the countries chosen for IMF programs (Przeworski and Vreeland 2001). We argue here that there are good theoretical reasons to side with the latter camp. First, IFIs threats to discontinue lending to non-conforming countries simply are not credible. This argument has been demonstrated formally. Because developing-country government leaders understand that part of the motivation behind aid is poverty alleviation, they also understand that donors will disburse future money even if recipients fail to follow through on attached conditions. Therefore, commitment fails and conditionality is feckless (Svensson 2000). Relatedly, international financial institutions were created for the purpose of engendering stability in the global economy and moving large sums of money to do so. If IFIs do not move money, they lose their purpose. Therefore, recipient governments understand that IFIs must lend to them even if their records on conditionality are spotted. Commitment fails doubly. To these well-understood reasons for the failure of conditionality, we add an additional reason that has largely been overlooked: the recipient countries have significant voice in the decisionmaking of the IFIs. The IMF, the World Bank, and all of the regional development banks are governed like joint stock corporations, with member governments as shareholders. Together, the member governments comprise a collective principal that governs the agents composed of the IFIs staffs. In agency theory,

6 principals and agents never exactly coincide, but principals have information-gathering and contracting and tools by which they can monitor and motivate agents to pursue the principals welfare. Self-interested agents, of course, can hide information and their actions from principals and can even use the authority delegated them against the principals interests (see Kiewiet and McCubbins 1990; Hawkins et al. 2006). International organization adds a twist to agency theory: the principal is not a single individual but a corporate entity comprised of many members (Nielson and Tierney 2003; Lyne et al. 2006). In order for a collective principal to contract and re-contract with an agent, a critical mass (often a majority) of members must coalesce into a bloc that is empowered to act for the whole. In the case of the IFIs, a majority vote of the shareholders is required to approve all programs and projects and otherwise give marching orders to the IFI staffs. While voting shares at some IFIs are dominated by developed countries, developing countries most of them loan recipients have significant shares as well. At the World Bank and the European Bank for Reconstruction and Development these developingcountry shares total nearly fifty percent of all shares. At the IMF, voting rules weight developing countries even more. And at the regional development banks such as the Asian, Inter-American, and African Development Banks developing countries often hold strong majority shares. Evidence suggests that the combined voting weights of the developing world have made critical differences in the recent changes in emphases at the development banks, particularly in causing the large increases in social lending for health, education and safety nets (see Lyne, Nielson and Tierney 2006).

7 Applied to adjustment lending, the strong voting weights and potential for collusive coalition-building of recipient countries likely means that commitments to enforce conditionality are not credible because of a systemic conflict of interest. The very countries subject to the threats to cut off financing are themselves non-trivial members of the collective principals that govern the IFIs and ostensibly enforce the commitments. As game theory emphasizes, promises to self enforce are usually cheap talk. This additional rationale for the toothlessness of conditionality only underscores Mallaby s conclusion that conditionality is merely mutually assured hypocrisy (Mallaby 2004, 46). Thus, as a central hypothesis, we do not expect that receipt of adjustment loans will make a significant difference in governments subsequent policy choices. Methods and Results We obtained our data from the PLAID database, a collaboration between the College of William and Mary and Brigham Young University that seeks to compile all development finance statistics, both bilateral and multilateral, during the period 1970 to 2003 (PLAID, 2006). We extracted all structural adjustment loans during the period from six multilateral aid agencies (the African Development Bank, the Asian Development Bank, the Inter-American Development Bank, the European Union, the World Bank, and the United Nations Development Programme) and five of their bilateral counterparts (the United States, France, the United Kingdom, Germany, and Japan.) We obtained data from the International Monetary Fund from their website (IMF, 2006). Using these project-level data, we obtained the total value of structural adjustment loans each recipient nation received from all donors in each year (in 2000 US, as well

8 as the total from each individual multilateral or bilateral institution. Additionally, we obtained the total number of projects each nation received during the period, as well as the total number from each of the respective institutions. We obtained our controls from the Development Research Institute at New York University, which compiles data from the World Bank s World Development Indicators and Global Development Finance databases (World Bank, 2006). We include three different measures of structural adjustment loans. For the first, we include only those loans that are specifically tagged as structural adjustment projects by the donor institutions. For the second, we include the first measure and add balance of payments and budget support projects. The third measure includes the first two and adds import support. We use these three measures because all of these projects are given to countries in economic crises, but the policy conditions between the three are not clear. The first measure is the most stringent and will be referred to throughout the text, but all analyses were performed using each of the three measures. Using the different measures did not yield significantly different results, which are therefore not reported. Table A Projects Countries Projects Countries >

9 Table A gives a sense of the distribution of the data. Sixty-nine countries received no structural adjustment packages between 1973 and 2000, whereas eleven received between fifty and seventy-five. Roughly half of the countries in the study received five loans or fewer. (To see the distributions for the second and third measure of structural adjustment, please refer to table C and D, respectively, in the appendix.) Table B stats log inflation log ex rate deficit SA total proj count min mean E median max E sd E AFDB_1 ASDB_1 EU_1 IADB_1 IMF_1 min E mean median max 9.71E E E+07 sd UNDP_1 WB France_1 min mean median max E E+08 sd E Table B gives the summary statistics for the relevant data. As a first cut in looking at basic relationships in the data, we opted for a set of pooled time-series cross-sectional regressions assessing the effects of adjustment loans on overall levels of three policy variables: inflation, budget deficits, and exchange rate stability. As an initial approach to this question, we begin by using prais-winsten regressions to test whether structural adjustment projects have a significant effect on our

10 three policy variables. All policy variables are differences to allow the model to address changes in policy rather than static levels. All analyses are performed according to the following hypothesize relationships: policy it X it SA _ dollars it SA _ project _ count it recipient i it policy it X it SA _ dollars ijt SA _ project _ count ijt recipient i it policy it X it SA ijt recipient i it where i indicates recipient, t represents year, j represents donor, X represents a matrix of covariates, recipient i indicates a vector of dummy variables for each recipient i, and SA represents a vector of dummy variables for each recipient donor pair, indicating whether or not country i received a structural adjustment project from donor j in year t. Using Prais-Winsten regressions with panel-corrected standard errors and an AR1 correction, we attempted to determine whether the amount of structural adjustment dollars, the number of structural adjustment loans, or the mere presence of a structural adjustment project cause nations to alter their economic policy (Prais and Winsten,1954; Beck and Katz 1995). We selected the Prais-Winsten model to control for possible autocorrelation across time. We lagged each of our key independent variables one period to test for possible implementation lags. As controls, we include the exchange rate (except where exchange rate stability is the dependent variable), population growth, GDP (measured in US, GDP growth, and external debt, in addition to recipient fixed effect dummies to control for unit effects. We began by using inflation as our dependent variable. Our results indicate that only bilateral projects from Japan seem to have an impact on inflation, with a one million dollar adjustment project corresponding to a 4.89 decrease in the rate of inflation (i.e.

11 inflation would slow down by 4.89%. Note this does not indicate a reversal, merely a slowing.) Neither the size of total adjustment funds nor the total number of projects are significant. It is important to note that the signs are in the expected direction, but that the standard errors indicate that there is no conclusive evidence that structural adjustment projects have any impact on inflation in recipient nations (See table E in the appendix.) When examining exchange rate stability, it appears that structural adjustment projects do not have a statistically significant effect, regardless of donor, size of loan, or number of loans (see table F in the appendix). Our third dependent variable is the change in budget deficits. These results indicate that structural adjustment projects, do, in fact, decrease the change in budget deficits. Projects from the EU, IMF, World Bank, and UK all have an impact on deficits. In each case, the signs on the coefficients for dollars and the coefficients for the number of projects are in the opposite direction, indicating an offsetting effect. In total, however, these effects are negative. These final results serve as a good control for our theory. Budget deficits are the most proximate of our three variables and would most likely be directly affected by structural adjustment projects. In some ways, the very act of receiving a project should directly impact the deficit. The fact that deficits are affected by adjustment projects while neither inflation nor exchange rates are indicates that recipients, may, in fact, be opting to avoid policy changes in those areas they have direct control over. Conclusion

12 We began with the hypothesis that we expected little effect of structural adjustment programs on subsequent government policy. The reported results suggest a good deal of supportive evidence for this argument. Changes in the overall levels of policy indicators pose only the easiest test for the effects of structural adjustment, however. This paper has examined only whether structural adjustment projects have dampened economic crises, let alone reversed them. Given that limited evidence exists that adjustment loans subsequently induce the expected policy changes, it is likely that when we raise the threshold for policy reform that the effects will diminish even further. A promising avenue would be to perform hazard analysis of the effects of adjustment packages on the rate at which governments hit specified targets for policies, such as designated levels of inflation, budget deficits, and exchange rate stability. Overall, adding a much larger portion of the universe of adjustment lending and testing the more proximate effects of that lending on policy variables seem promising directions. If the results here are borne out with additional methods and testing, it should help policymakers considering the reform of conditionality practices and discretional financing to focus on more effective methods of promoting the well being of citizens in member countries.

13 References Beck, Nathaniel, and Jonathan N. Katz What To Do (and Not To Do) with Times- Series Cross-Section Data. American Political Science Review 89: Burnside, C. and D. Dollar Aid, policies, and growth. American Economic Review 90, 4: Hansen, Henrik and Finn Tarp Aid and growth regressions. Journal of Development Economics. 64, 2: International Monetary Fund IMF Lending Arrangements. At < org/external/np/tre/tad/extarr1.cfm.>. Accessed 14 August 2006 Kiewiet, D. Roderick, and Mathew D. McCubbins The Logic of Delegation: Congressional Parties and the Appropriations Process. Chicago: University of Chicago Press. Lyne, Mona, Daniel L. Nielson, and Michael J. Tierney Who delegates? Alternative models of principals in development aid. In Darren Hawkins, David Lake, Daniel Nielson, and Michael Tierney, editors. Delegation and Agency in International Organizations. New York: Cambridge University Press. Mallaby, Sebastian The world s banker. New York: Penguin Press. Morrissey, Oliver Conditionality and aid effectiveness re-evaluated. World Economy. 27, 2 : Noorbakhsh, F. and A. Paloni Structural adjustment and growth on Sub-Saharan Africa: The importance of complying with conditionality. Economic Development and Cultural Change 49, 3: Nielson, Daniel L., and Michael J. Tierney Delegation to international organizations: Agency theory and World Bank environmental reform. International Organization 57: PLAID At < Prais, S. J., and C. B. Winsten Trend Estimators and Serial Correlation. Cowles Foundation, Discussion Paper 383, Chicago. Przeworski, Adam and James R. Vreeland The effect of IMF programs on economic growth. Journal of Development Economics. 62: Stiglitz, Joseph What I learned at the world economic crisis. New Republic.

14 Svensson, Jakob When is foreign aid policy credible? Aid dependence and conditionality. Journal of Development Economics. 61: World Bank World Development Indicators. At < /dataonline/> Global Development Finance. At < apponline/default.asp?app=gf Appendix Table C Projects Countries >75 Projects Countries Table D Projects Countries >75 Projects Countries

15 Table E Inflation (1) (2) (3) Variable inflation inflation inflation Totals SA dollars ( ) # of Projects (14.0) Totals by Donor Donor dummies SA dollars -1.75e-09 (lagged 1 period) ( ) # of Projects (lagged 1 period) (15.4) AFDB dollars ( ) ASDB dollars 0 EU dollars ( ) IADB dollars 0 IMF dollars ( ) UNDP dollars (0.077) WB dollars ( ) France dollars ( ) Germany dollars ( ) Japan dollars *** ( ) UK dollars ( )

16 US dollars ( ) AFDB count (87.1) ASDB count 0 EU count (13.1) IADB count 0 IMF count (63.9) UNDP count (378) WB count (79.5) France count (28.7) Germany count (159) Japan count (102) UK count (65.6) US count (478) IMF*WB -0 IMF count*wb count (96.6) AFDB dollars (lagged 1 period) ( ) ASDB dollars 0 (lagged 1 period) EU dollars (lagged 1 period) ( ) IADB dollars 0 (lagged 1 period)

17 IMF dollars (lagged 1 period) ( ) UNDP dollars (lagged 1 period) ( ) WB dollars (lagged 1 period) ( ) France dollars (lagged 1 period) ( ) Germany dollars (lagged 1 period) ( ) Japan dollars (lagged 1 period) ( ) UK dollars (lagged 1 period) ( ) US dollars (lagged 1 period) ( ) AFDB count (lagged 1 period) (93.6) ASDB count 0 (lagged 1 period) EU count (lagged 1 period) (11.4) IADB count 0 (lagged 1 period) IMF count (lagged 1 period) (59.9) UNDP count (lagged 1 period) (106) WB count (lagged 1 period) (77.2) France count (lagged 1 period) (35.2) Germany count (lagged 1 period) (181) Japan count (lagged 1 period) (102) UK count (lagged 1 period) (95.3) US count

18 (lagged 1 period) (402) IMF (66.1) EU (48.0) AFDB (54.1) IADB 0 UNDP (115) WorldBank (101) France (42.8) Germany (186) Japan (108) UK (105) US (442) Bretton_Woods (209) L.IMF (lagged 1 period) (63.7) L.EU (lagged 1 period) (60.3) L.AFDB (lagged 1 period) (61.7) L.IADB 0 (lagged 1 period) L.UNDP (lagged 1 period) (69.8) L.WorldBank (lagged 1 period) (103) L.France (lagged 1 period) (52.0)

19 L.Germany (lagged 1 period) (200) L.Japan (lagged 1 period) (108) L.UK (lagged 1 period) (120) L.US (lagged 1 period) (416) Exchange Rate Population Growth GDP Growth * * * GDP 1.2E E E E E E-10 External Debt -3.09E E E E E E-10 Constant (51.1) (81.0) (95.7) Observations countryid R-squared Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Fixed effects not reported Table F Exchange Rate Stability (1) (2) (3) Variable Exchange Rate Totals SA dollars Exchange Rate Totals by Donor Exchange Rate Donor dummies

20 ( ) # of Projects (684) SA dollars (lagged 1 period) ( ) # of Projects (lagged 1 period) (664) AFDB dollars ( ) ASDB dollars 0 EU dollars ( ) IADB dollars 0 IMF dollars (0.0052) UNDP dollars (34.0) WB dollars ( ) France dollars ( ) Germany dollars ( ) Japan dollars ( ) UK dollars ( ) US dollars ( ) AFDB count (2123) ASDB count 0 EU count

21 (499) IADB count 0 IMF count (3989) UNDP count (163357) WB count (8448) France count (973) Germany count (3430) Japan count (1546) UK count (1427) US count (2694) IMF*WB 0 IMF count*wb count (2681) AFDB dollars (lagged 1 period) ( ) ASDB dollars 0 (lagged 1 period) EU dollars (lagged 1 period) ( ) IADB dollars 0 (lagged 1 period) IMF dollars (lagged 1 period) (0.0063) UNDP dollars (lagged 1 period) (0.17) WB dollars (lagged 1 period) ( ) France dollars (lagged 1 period) ( )

22 Germany dollars (lagged 1 period) ( ) Japan dollars (lagged 1 period) ( ) UK dollars (lagged 1 period) (0.0024) US dollars (lagged 1 period) ( ) AFDB count (lagged 1 period) (5518) ASDB count 0 (lagged 1 period) EU count (lagged 1 period) (581) IADB count 0 (lagged 1 period) IMF count 4498 (lagged 1 period) (4052) UNDP count (lagged 1 period) (23364) WB count (lagged 1 period) (2038) France count (lagged 1 period) (1217) Germany count (lagged 1 period) (4804) Japan count (lagged 1 period) (2452) UK count 2033 (lagged 1 period) (18105) US count (lagged 1 period) (2868) IMF (4509) EU (1018) AFDB (4354) IADB 0

23 UNDP 1890 (7943) WorldBank (13093) France (1840) Germany (3454) Japan (1748) UK 3533 (12883) US (2407) Bretton_Woods (17846) L.IMF 5133 (lagged 1 period) (4582) L.EU (lagged 1 period) (812) L.AFDB (lagged 1 period) (3056) L.IADB 0 (lagged 1 period) L.UNDP 2184 (lagged 1 period) (5493) L.WorldBank (lagged 1 period) (9896) L.France (lagged 1 period) (2134) L.Germany (lagged 1 period) (3778) L.Japan (lagged 1 period) (1729) L.UK (lagged 1 period) (14142) L.US (lagged 1 period) (1589)

24 Population Growth (738) (659) (699) GDP Growth (267) (245) (239) GDP ** * ** ( ) ( ) ( ) External Debt ( ) ( ) ( ) Constant (3178) (2880) (2856) Observations countryid R-squared Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Fixed effects not reported Table G Budget Deficits (1) (2) (3) Variable Deficit/Surplus Deficit/Surplus Deficit/Surplus Totals Totals by Donor Donor dummies LD.ovrall_def_surp (0.22) (0.21) (0.22) SA dollars ( ) # of Projects 3674 (2447) # of Projects -5968** (lagged 1 period) (2496) AFDB dollars

25 ( ) ASDB dollars 0 EU dollars ** ( ) IADB dollars 0 IMF dollars ** ( ) UNDP dollars (3.27) WB dollars * ( ) France dollars ( ) Germany dollars (0.0012) Japan dollars ( ) UK dollars ** ( ) US dollars ( ) AFDB count (33044) ASDB count 0 EU count -7982*** (2940) IADB count 0 IMF count 20009*** (7529) UNDP count 0 WB count 20186**

26 (10088) France count 7356 (10597) Germany count (43478) Japan count 2644 (13070) UK count ** (12188) US count 4098 (40142) IMF*WB 0 IMF count*wb count (15115) AFDB dollars (lagged 1 period) ( ) ASDB dollars 0 (lagged 1 period) EU dollars * (lagged 1 period) ( ) IADB dollars 0 (lagged 1 period) IMF dollars * (lagged 1 period) ( ) UNDP dollars (lagged 1 period) (3.08) WB dollars ** (lagged 1 period) ( ) France dollars ** (lagged 1 period) ( ) Germany dollars (lagged 1 period) ( ) Japan dollars (lagged 1 period) ( ) UK dollars (lagged 1 period) ( ) US dollars (lagged 1 period) ( )

27 AFDB count 48597* (lagged 1 period) (25700) ASDB count 0 (lagged 1 period) EU count 5200** (lagged 1 period) (2549) IADB count 0 (lagged 1 period) IMF count * (lagged 1 period) (6192) UNDP count 0 (lagged 1 period) WB count ** (lagged 1 period) (18855) France count (lagged 1 period) (10195) Germany count (lagged 1 period) (22333) Japan count 6502 (lagged 1 period) (22799) UK count 2272 (lagged 1 period) (10520) US count (lagged 1 period) (29984) IMF 13595* (7883) EU 30560** (14408) AFDB (10376) IADB 0 UNDP (11468) WorldBank 11210** (5404) France (6884) Germany 1951

28 (18399) Japan 1454 (14908) UK 3349 (4635) US 2999 (46485) Bretton_Woods (18055) L.IMF (lagged 1 period) (5905) L.EU ** (lagged 1 period) (17004) L.AFDB 19732** (lagged 1 period) (9364) L.IADB 0 (lagged 1 period) L.UNDP (lagged 1 period) (8793) L.WorldBank ** (lagged 1 period) (9516) L.France 14685* (lagged 1 period) (8469) L.Germany 8284 (lagged 1 period) (12577) L.Japan (lagged 1 period) (20663) L.UK (lagged 1 period) (5274) L.US (lagged 1 period) (45430) Exchange Rate Population Growth GDP Growth GDP -2.21E E E E E-09

29 External Debt -6.25E E E Constant (5462) (5797) (5970) Observations countryid R-squared Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Fixed effects not reported