Data Collection Analysis Technique

l Figure 1 Theoretical Framew orks

CHAPTER III RESEARCH M ETHOD

A. Data Collection

Dat a is collect ed f rom t he published audit ed financial st at ement s from t he official w ebsit e of BPK f rom year 2006 t o 2007. The published financial st at ement s from local government s w ill include several t ypes of audit opinion excluding adver se and disclaimer. Dat a for macroeconomic fact ors ar e collect ed f rom various sources especially dat a available in w ebsit es provided by Indonesia St at ist ics Agency or BPS , Bank of Indonesia BI and many ot her support ive w ebsit es.

B. Population and Sample

li

1. Population

Populat ion ref ers t o t he ent ire group of people, event s, or t hings of int er est t hat t he researcher w ishes t o invest igat e Sekaran, 2003. The populat ion of t his research is all published audit ed financial st at em ent s of Indonesian local governm ent s f or t w o years successively f rom 2006 t o 2007.

2. Sample

A sample is a subset of populat ion and it comprises some mem bers select ed fr om t he populat ion Sekaran, 2003. This research uses judgment al sampling t o draw t he samples fr om financial st at ement s. The use of judgment al sampling aimed t o achieve t he samples by choosing subject s t hat are in t he most suit able place t o give specif ied informat ion Sekaran, 2003. All published financial st at em ent s cont aining any t ypes of opinion except adverse and disclaimer w ill be included in dat a calculation.

C. Analysis Technique

This research w ill use mult ivariat e r egr ession analysis t hat is done by using ordinary least squares OLS. Ordinary least squar es OLS is a t echnique for est imat ing t he unknow n paramet ers in a linear regr ession model w hich minimizes t he sum of squared dist ances bet w een t he obser ved responses in a set of dat a, and t he fit t ed responses from t he r egr ession model w w w .w ikipedia.org. M ult ivariat e r egression analysis is used w hen t her e are t w o or mor e dependent variables exist in t he model Khat t r ee and Naik, 2003. Because of t he nat ur e of t he dat a t hat non-linear in nat ure, t he variables such as populat ion, inflat ion and GRDP w ill be t ransformed int o logarit hmic forms Chat t erj ee and Hadi, 2006. lii The model w ill be proposed as follow s: Financial Rat ios= a0 + b1 Log GRDP + b2 Log Pop + b3 Log Infl + b4 M undic + b5 Cireg Where; Log GRDP = Logarit hmic f orm of Gross Regional Dom est ic Product Log Pop = Logarit hmic f orm of Populat ion Log Infl = Logarit hmic f orm of Inflat ion IBLOG = Island-based Locat ion of Local Governm ent s ASLOG = Administ rat ive St at us of Local Governm ent s To comput e t he regression model, t his research w ill use GRETL Version 1.1 Soft w are. The assumpt ion of t he models are as following:  Nonlinearit y t est : This kind of t est is used t o check t he linearit y of t he r egression model. The model of linear regression must be linear.  Het eroskedast icit y: Her oskedast icit y happens w hen random variables have differ ent variances. Het eroskedast icit y problem of t en occurr ed in OLS t echnique. The assumpt ion of homoskedast icit y has t o be maint ained. How ever, if het eroskedast icit y is occurr ed or t he homoskedast icit y is violat ed, t hen anot her st at ist ics t echnique has t o be conduct ed. The alt ernat ive w ay t o OLS w hen homoskedast icit y exist s is w eight ed least squar es Adkins, 2007. liii The w eight ed least squar es t echnique in Gret l is conduct ed using het er oskedast icit y correct ed t est .  Collinearit y: This t est is aim ed t o indicat e t he degr ee t o w hich each of t he explanat or y variables is collinear w it h t he ot her explanat or y var iables. When t his sit uat ion exist s, t he model suf fer s mult icollinearit y problems. The good regr ession model must f r ee from t his problem. In Gret l, t his assumpt ion is checked by indicat ing variance inf lat ion fact ors VIF value. When t he VIF is higher t han 10, t hen t he model suffers mult icollinearit y problem.  Normalit y: Nonnormalit y of errors oft en r equir ed by r egr ession model. In t he cont rary, ordinary least squar es OLS t echnique does not require t his assumpt ion Ramanat han, 1993. As st at ed by Ramanat han 1993, t he most impor t ant t hing in OLS is t hat t he dat a is nonsingular t ype free of het eroskedast icit y and mult icollinearit y problems.

D. Variable Identification