The Simple Fixed Effects

Firstly, the effect o f changes in minimum wage on average wage is estimated using the simple fixed effects method. Table 5.3 presents a set o f simple fixed effects method results using four different measures o f minimum wage. As presented in the first column, the coefficient o f log o f real minimum wage is estimated to be positive and significant at 1% level. Using the log of real minimum wage, it is suggested that a 10% increase in the real minimum wage raises the real average monthly wage by 1.66%. The estimated coefficient is relatively higher than Rama’s (2001) which use the same measure in the previous Indonesian minimum wage study. In comparison, Rama (2001) found that an increase in the minimum wage by 10% increased the average wage by 1 %.

Compared to Rama (2001), firstly, there is a difference in the sample period as this study uses a more recent panel data set. Secondly, relating to the control variables, Rama (2001) used urban population and the Gross Domestic Product as the control variables, while some more specific demand and supply shifters are employed in this

18 T o u g h n e s s is e x c lu d e d in th e w a g e e q u a tio n as a re su lt o f th e p o te n tia l c o rre la tio n w ith th e w a g e rate in th e le ft-h a n d s id e -e s tim a to r (se e B u rk h a u s e r et al. 2 0 0 0 ).

study. Thirdly, Rama (2001) focused on the effects o f minimum wage on average wage o f full-time workers in urban areas and in the manufacturing sector, while the effects o f the minimum wage on the average wage o f total paid employment is discussed in this study, including the average wage o f part-time workers that are also legally covered by the minimum wage policy.

In the second column, the minimum wage is measured by the fraction affected which comprises the proportion o f workers that earn a wage between the old and new level o f minimum wage. Similar to the log o f real minimum wage, the fraction affected coefficient is positive and significant. In comparison, the coefficient o f the fraction affected is relatively higher than the log o f real minimum wage with the coefficient of 0.293. However, as pointed out by Lemos (2005), the coefficient o f the fraction affected is not directly comparable to the log o f real minimum wage coefficient because they measure the minimum wage effect in a different way. In practice, the log o f the real minimum wage measures the level o f the minimum wage, while the fraction affected measures the proportion o f workers affected by the minimum wage.

To make it comparable with the log o f real minimum wage coefficient, as pointed out by Lemos (2005), the fraction affected coefficient needs to be multiplied with the elasticity o f the fraction affected with respect to the log o f real minimum wage19. The result shows that an increase in minimum wage by 10% raises the fraction affected by 3.73%, suggesting an increase in average wage by 1.09%. In other words, the result indicates that the “comparable” or adjusted coefficient o f the fraction affected (0.109) is slightly lower than the log o f real minimum wage coefficient (0.166).

19 S p e c ific a lly , th e fra c tio n a ffe c te d e la s tic ity is o b ta in e d b y e s tim a tin g th e fra c tio n a ffe c te d v a ria b le on th e lo g o f real m in im u m w a g e an d all o f the c o n tro l v a ria b le s.

In the next two columns, the minimum wage is measured by the fraction at and the fraction below, which also measures the degree o f impact o f the minimum wage.

Unlike the previous measures, the fraction at and the fraction below estimates show inconsistent results with what we would expect, suggesting negative effects o f the minimum wage on the average wage. This contrary result can be explained by writing that a rightward shift o f the earnings distribution automatically raises the average wage and reduces the fraction at and the fraction below measures, suggesting a potential endogeneity bias.

In addition, as noted above, these minimum wage measures might not be effective for the Indonesian case for a number o f reasons. Firstly, the fraction at is potentially underestimated to measure the degree o f impact o f minimum wage in Indonesia because the minimum wage can only be applied to the workers who work for less than one year20. It seems that the Indonesian minimum wage tends to be an instrument for raising the standard o f living o f workers rather than exactly as a wage floor (Manning, 2003a). Secondly, Lemos (2004d) also pointed out that the fraction at is potentially endogenously determined with the labour market condition because the wage bargaining will decide which workers are paid at (or around) the minimum wage level, suggesting a potential endogeneity biased estimate. Specifically, as pointed out by Lemos (2004d), the fraction at will increase when the minimum wage increases and then will decrease when workers who are previously not paid below the minimum wage level bargain to increase their wages above the minimum wage level. Further details about the endogeneity problem will be discussed in the next section. On the other hand, the fraction below measure potentially includes workers (paid

20 B a se d on g o v e rn m e n t re g u la tio n , m in im u m w a g e w o rk e rs w ith o n e y e a r d u ra tio n o f w o rk m u st be p a id a b o v e th e m in im u m w a g e lev el, a lth o u g h it is n o t c le a r by how' m u ch .

employment) not truly affected by the minimum wage because o f a large non- compliance problem and the greater proportion o f workers in the uncovered sector in Indonesia, suggesting a non-effective minimum wage measure for Indonesia.

Table 5.3 Wage Equation using the Simple Fixed Effects Estimate

Log of Real

Minimum Wage Fraction Affected Fraction At Fraction Below

(1) (2) (3) (4)

Coef. P value Coef. P value Coef. P value Coef. P v a lu e MW measure 0.1656 0.000 0.2935 0.021 -0.9075 0.019 -0.1599 0 .0 8 7

Urban 0.7044 0.000 0.3446 0.021 0.6432 0.000 0.6422 0.000

Youth -0.2460 0.542 -0.4899 0.263 -0.4243 0.298 -0.4845 0 .2 3 9

Women 0.4043 0.294 -0.4727 0.319 0.3434 0.381 0.4427 0.261

Industry -0.9629 0.015 -0.6700 0.124 -0.8183 0.042 -0.8556 0 .0 3 4

Trade -1.4901 0.000 -0.8123 0.056 -1.3792 0.000 -1.3426 0.001

Services -0.1874 0.606 -0.2056 0.632 -0.2392 0.517 -0.1591 0 .6 6 7 Construction 0.5543 0.511 -0.6238 0.484 0.6990 0.414 0.7852 0 .3 6 2 High School -0.0818 0.773 0.4359 0.126 -0.1790 0.534 -0.1655 0 .5 6 7

University 1.0618 0.175 3.4082 0.000 1.3898 0.081 1.2771 0 .1 0 9

Unemp Rate (-1) -0.8432 0.057 -0.6435 0.218 -0.9136 0.042 -0.9514 0 .0 3 5

Year 1990 0.0800 0.006 0.0939 0.002 0.1023 0.001

Year 1991 0.0629 0.060 -0.0823 0.034 0.1323 0.000 0.1403 0.000

Year 1992 0.0977 0.004 -0.0392 0.303 0.1770 0.000 0.1833 0.000

Year 1993 0.1950 0.000 0.0077 0.841 0.2915 0.000 0.3003 0.000

Year 1994 0.1925 0.000 -0.0859 0.029 0.3230 0.000 0.3414 0.000

Year 1995 0.2293 0.000 -0.0137 0.779 0.3754 0.000 0.3944 0.000

Year 1996 0.2731 0.000 -0.0821 0.093 0.4225 0.000 0.4389 0.000

Year 1997 0.3373 0.000 -0.0295 0.450 0.4920 0.000 0.5039 0.000

Year 1998 0.1428 0.002 -0.3399 0.000 0.2452 0.000 0.2540 0.000

Year 1999 0.2154 0.000 -0.0402 0.304 0.3069 0.000 0.3148 0.000

Year 2000 0.3451 0.000 0.0629 0.090 0.4613 0.000 0.4730 0.000

Year 2001 0.4775 0.000 0.0428 0.295 0.6249 0.000 0.6325 0 .0 0 0

Year 2002 0.4597 0.000 -0.1030 0.007 0.6254 0.000 0.6374 0.000

W ald test is used for group-w ise heteroscedasticity with the null hypothesis o f hom oscedasticity.

W ooldridge test is used for first-order autocorrelation test with the null hypothesis o f no first order autocorrelation.

In order to check the robustness o f these results, the diagnostic tests for serial correlation and heteroscedasticity are undertaken for the simple fixed effects estimates. The Wooldridge serial correlation test provides information about evidence o f serial correlation in the error term over time within provinces. If the null hypothesis is rejected, it would imply that there is evidence o f serial correlation in the estimation.

Moreover, a Wald test o f heteroscedasticity is also undertaken to examine whether the panel residuals change over time. The simple fixed effects model assumes the estimation to be homoscedastic in the regression residuals across provinces and years.

Ignoring both o f these issues would make our estimates inefficient. The estimated standard errors o f the regression coefficient would also be biased and inconsistent.

The diagnostic test results for serial correlation and heteroscedasticity are presented in the lower part o f table 5.3. The p-values for the Wooldridge test are significant at 5%

level suggesting that there is no evidence o f serial correlation using the simple fixed estimate. However, according to Wald test results, all o f the minimum wage measures suffer from heteroscedasticity. Ignoring this issue will make our estimates inefficient.

Therefore, we argue that the results o f the simple fixed effect estimates are not robust enough in examining the effect o f minimum wage on the average wage in Indonesia.