PARTIAL LEAST SQUARES METHOD TO ANALYZE ECONOMIC GROWTH
IN KALIMANTAN ISLAND
BASE ON DESKTOP APPLICATION

Melvin Hendri, Rokhana Dwi Bekti, Franky H. Marpaung
Universitas Bina Nusantara,
Jl. Syahdan No. 9, Kebon Jeruk, Jakarta Barat 11480, 021-5345830

ABSTRACT

Structural Equation Model (SEM) is a multivariate statistics analysis which used for construct, estimate, and examine causal relationship. The object of this research are apply Partial Least Square (PLS) estimate method for analyze SEM and develop an efficient desktop application to analyze PLS method. PLS method is used as an alternative method for analyzing SEM model which have small number of data and non-normal distributed. This research use R software to help researcher analyze with PLS method. For desktop application development, this research use Java language and Net Beans software. Topic of this research is influence of natural and human resources for economic growth at Kalimantan Island. PLS analyze method which used in this research are validity, reliability, R-squared, and significant test. The result of this research show that agriculture production and labor factor not meet validity assumption. After respecification, this research get result that all manifest variable meet validity and reliability assumption. R2 score of this research is 51%, it mean that the model which developed is a good model. For significant test, this research use significant level 5%, result of this research show that natural and natural resources have a significant influence for economic growth at Kalimantan Island.

Keyword: Partial Least Square, natural resources, human resources, economic growth, desktop application

Introduction

Structural Equation Modeling ( SEM ) is a method which is formed due to the problem of measuring a variable that can not be measured directly (Santoso , 2012: 1). Variables that can not be measured is called as latent variable which requires a manifest variable as an indicator or measurement of the latent variables . During its development , SEM become a popular method because it can be applied to several analyzes , such as analysis of causal modeling , confirmatiory analysis, second order factor analysis , regression models analysis, analysis of covariance structure models , structure models and correlation analysis (Mustafa and Wijaya , 2012: 4 -5).

There are several methods of estimation in the SEM method , but all of the SEM estimation methods have drawbacks, which requires large amounts of samples, can not have multicollinearity and data that should be normally distributed . Therefore , developed an alternate method for estimating SEM that aims to overcomes the shortcomings in the other methods, the method is Partial Least Square (PLS). Partial Least Square (PLS) is one of the alternate methods of estimation models for managing Structural Equation Modelling (SEM). PLS design created to overcome the limitations of SEM methods. In the SEM method requires huge amounts of data , there are no missing values ​​, should be normally distributed , and should not have multicollinearity , whereas the PLS using a distribution free approach where certain data can be distributed. Moreover PLS can also be used on small sample sizes. (Mustafa and Wijaya , 2012: 11).

In this study raised the topic about the influence of Natural Resources (NR) and the Human Resources (HR) on economic growth. In the time of globalization, economic conditions is one of the aspects that are considered important in the social life. According to Todaro and Smith (2006) in Indrasari (2011: 19), economic development is a multidimensional process that includes a variety of fundamental changes in the social structure, the attitudes of the people, and national institutions, in addition to still pursue economic growth acceleration, handling income inequality, and poverty eradication. Thus if one country have a high economic growth, this country also havethe high the welfare of the community. Therefore, economic growth can be used as a reference to assess the level of success of a government in improving the welfare of society.

In this study, the variables usedareeconomicgrowth, natural resourcesand human resources. Thethirdvariableis acomplexvariableand can not bemeasured directly. Becauseour modelinvolvesseverallatentvariablesandindicatorsneed to beanalyzed, it must bemodeled aStructural Equation Model (SEM). Toovercome the limitations ofSEMshortageinthe numberanddistributionof data, thenthis study usingPartial LeastSquare(PLS).

Alongwith theadvancement ofinformationtechnology, there are somespecial softwaretoanalyze theSEMmodelssuch asLISREL, AMOS, andR.By usingthis software, acomplexmodel thatcan be tested, eitherthe relationships between manifestvariables(indicators) withlatent variables, nor relationships amonglatentvariables. At this time, allmodels ofSEManalysis toolsdo notfocusonanalyzing theSEMmodelsbutalsohas the function ofanotherapplication, therefore theresearchers builtadesktopapplicationthat is usedspecificallyto analyze theSEMmodelsthat can be understoodbythe userin analyzingandmakingconclusionsmodel analysisSEM.

Methods

In this study,usingsecondary data, where the dataobtainedfromthe dataalreadyavailable. Secondary datainthis study were obtainedfrom theIndonesianCentral Statistics Agency(BPS). The data usedis the datain 2009. The amount of datausedby 52districtson the Kalimantan Island
Step method of analysisin this studyare:

  1. Performdescriptive statisticsforeachvariableasthe calculation ofthe mean, variance, median, standarddeviation, and others.
  2. Designingandconvert themeasurement model(outermodel)

Fishery = λ11 NR + ε1

Plantation = λ21 NR + ε2

Agriculture = λ31 NR + ε3

Labor Force = λ12 HR + ε4

Literacy = λ22 HR + ε5

GDP = λ13 ECONOMIC + ε6

Expenditures = λ23 ECONOMIC + ε7 (1.1)

  1. Designingandconvertingthe structural model(innermodel)

ECONOMIC = 11 NR + 12 HR + ζ1

  1. Construction ofthe path diagram

Path diagramin this studyare asfollows :

Picture 1. Path Diagram

  1. EstimatingModel
  2. EvaluatingOuterModel

The purposeofthe evaluationistoobtainmodelsoutermanifestvariablesarevalid and reliable. Evaluation ofthe validity oftheoutermodel can bedone in two ways, namelyconvergentvalidityanddiscriminantvalidity. Evaluationof reliability on theoutermodel can bedonewithcompositereliability

  1. EvaluatingInnerModel

The purposeofthe evaluationistodetermine theinnermodelsthatstructuralmodelscatergoriesas agood modelornot. Toevaluate theinnermodel oftheresearchis doneto test theR-squared (R2).

  1. Hypothesis Testing(resampling bootstrapping)
  2. Desktop-based application designusingUML
  3. The design ofthe applicationinterfaceby following thegolden rulesrule8(8 GoldenRules)

Results

There areseveralstatisticalanalyzesperformedin this study, namelythe descriptiveanalysis, validityanalysis, reliabilityanalysis, R2 test, SEMmodel buildingandtests of significance.

Validity Analysis

Validity analysis aimed to evaluate the validity of the manifest variables on the outer models. In the PLS method, the validity test can be done in two ways, namely convergent validity and discriminant validity.

In Table 1, it can be seen that the agricultural indicator and labor force indicator had loading less than 0.50. This suggests that these indicators do not meet the convergent validity.

Table 1. Loading Factor

Estimate
NR -> Fishery / 0,62
NR -> Plantation / 0,79
NR -> Agriculture / -0,38
NR -> Labor Force / -0,61
NR -> Literacy / 0,86
ECONOMIC -> Expenditures / 0,76
ECONOMIC -> GDP / 0,80

In Table2, it is seen that theagriculturaloutcome variableshas a greater correlation to theHR, not correlatedsignificantly to HR, so that agricultural products do not meet the discriminant validity. In addition to the variable work force has a greater correlation to the NR, so the variable labor force also did not meet the discriminant validity.

Table 2. Cross Loadings

NR / HR / ECONOMIC
Fishery / 0,620 / 0,240 / 0,280
Plantation / 0,791 / -0,036 / 0,265
Agriculture / -0,375 / -0,124 / -0,210
Labor Force / -0,302 / -0,610 / -0,397
Literacy / 0,031 / 0,859 / 0,616
Expenditures / 0,078 / 0,750 / 0,763
GDP / 0,556 / 0,352 / 0,804

Because it does notmeet theconvergentvalidityanddiscriminantvalidity, then the model wre modifiedby removing the agricultural variable and labor forcevariable. After themodification ofthemodel of, the new model will betested by convergentvalidityanddiscriminantvalidityagain.

Table 3. Loadings Factor after respecification

Estimate
NR -> Fishery / 0,86
NR -> Plantation / 0,71
NR -> Literacy / 1,00
ECONOMIC -> Expenditures / 0,83
ECONOMIC -> GDP / 0,73

From Table3, it can be seenthat thefactorloadingsoneachmanifestvariablehas aloadinggreaterthan 0,50. It shows thatallthe manifestvariablesthat it meets theconvergentvalidity.

Table4. CrossLoadingsafterrespecification

NR / HR / ECONOMIC
Fishery / 0,859 / 0,129 / 0,279
Plantation / 0,705 / -0,070 / 0,201
Agriculture / 0,056 / 1,000 / 0,662
Labor Force / 0,081 / 0,754 / 0,832
Literacy / 0,444 / 0,234 / 0,731

From Table4, it can be seenthat themanifestvariablesfishery andplantation cropshavethe highestcorrelationof thelatentvariables, ielatent variablesNR. Literacymanifestvariablealsohas thehighestcorrelationwith its latentvariables, namelyHRlatent variables. ManifestvariablesGDPandexpendituresalsohavethe highest correlationof the their latentvariables, it is latentvariableECONOMIC. This showsthatallthe variablesmanfiesmeetsvalidtyconvergent.

Reliability Analysis

Reliability analysisaimstoobtaina reliablelatentvariables. Intesting thereliability ofthePLSmodelscanuse thecompositereliability. Latent variablescan be said tohavehada goodreliabilityif thecompositereliabilityvalue is ​​greater than0.6.

Table 5. Composite Reliability

Composite Reliability
NR / 0,76
ECONOMIC / 0,76

OnTable5showsthatalllatentvariables NR andeconomicgrowthhascompositereliability0,6, so it canbe concludedthat the threelatentvariableshavegood reliability.

R-squared Test (R2)

R-squared (R2) test aimstodeterminehow well theinner workings ofthe model(structural model) areformed. AccordingGhozali(2011: 27), if theR2valueof 0.67; 0.33; 0.19forendogenouslatentvariablesin theinner model,itindicatesthat the model is"good", "moderate", and"weak". This studyhas avalue ofR-squared (R2) =0.51. This showsthatthisresearchhasa goodinnermodels.

SEM PLS Model

Modelsobtainedfromthe results ofthe studyare asfollows :

Picture 2. SEM PLS ModelJ

From Picture2, the equation of inner model obtained inthis studyare asfollows :

ECONOMIC = 0,27 NR + 0,65 HR

The equation of Outer modelsobtained inthis studyare asfollows :

Fishery = 0,86 NR

Plantation = 0,71 NR

Literacy = 1 HR

GDP = 0,83 ECONOMIC

Expenditures = 0,73 ECONOMIC

Tests of Significance

Tosee thesignificance ofeachvariablethere, thendothe testpath coefficientsestimatedthroughbootstrappingprocedure

Table 6. Path Coefficient

Estimate / Std Error / T Statistk
NR -> Fishery / 0,860 / 0,239 / 3,598
NR -> Plantation / 0,706 / 0,329 / 2,146
NR -> Literacy / 1,000 / 4,4 10-17 / 2,272 1016
ECONOMIC -> Expenditures / 0,833 / 0,103 / 8,087
ECONOMIC -> GDP / 0,731 / 0,060 / 12,183
NR -> ECONOMIC / 0,274 / 0,138 / 1,985
HR -> ECONOMIC / 0,647 / 0,062 / 10,435

Hypothesis T-1

Ho = There is no influence of NR on economic growth on the Kalimantan Island

H1 = There is the effect of NR on economic growth on the island of Borneo

In Table 6 shows that the model of the nr to economic growth have T statistic 1.985. This value is smaller than T α,df the 5% significance level is 1.960. Thus, H1 is accepted, that the NR has a significant influence on the economic growth on the Kalimantan Island.

Hypothesis T-2

Ho= There is no effect of HR on economic growth on Kalimantan Island

H1 = There is the influence of HR on economic growth on the Kalimantan Island

In Table 6 shows that the correlation between HR on economic growth statistics have T 10.435. Value is greater than Tα, df with a significance level of 5% is 1.960. Thus, H1 is accepted, that HR has a significant influence on the economic growth on the Kalimantan Island.

SpecificationsandApplicationProgram

Recomenned hardwarespecificationsforthisprogramare :

  1. Processor: Intel(R) Core(TM) i3-2350M cpu @2.3GHz
  2. Memory: 6GB
  3. Harddisk: 500GB
  4. VGA: Intel(R) HD Graphic Family (2GB)
  5. Monitor: Resolusi 1360 x 768

While therecommended softwarespecificationsforthisprogramare:

  1. NetBeans IDE version 7.3.1
  2. Java Development Kit (JDK) version 1.7.0_21
  3. R Language version 3.0.1. RLanguagesoftwareuseslibrarypackage, such aslibrary(Runiversal) whichis usedfor conversionRtoJavaandXML, library(XLConnect) is usedtoread thefilein xlsorxlsx, andlibrary(semPLS) is usedtoanalyzetheSEMmethodPLS.

Picture 3. Main Display ofApplications

The maindisplayof applicationis shownin Picture3. In this application, the usercanstartthe orderfromtop to bottom. In theImportmenuRscript, usercan importexistingRscript. Thenthe user can continuetobeusedto importdatausing theDataImportfunction. Then theusercan formlatentvariablesby using theAddLatentfunctions. User can also formthe manifestvariablesusing thefunctionAddManifestwhichmanifestvariables willfilledautomatically fromthe data-import. Thenusercanestablishrelationships ofstructuralandmeasurementrelationshipsusing thefunctionAddStructuralandAddMeasure. Once completedform the relationships, user can analyzingdata with GoToAnalyzefunction.

Apllication Evaluation

Applications are made has been evaluated by the Eight Golden Rule :

  1. Trying to be consistent

The interface is formed is made with one type of font. The font used is also the same size. In addition to the background color of the application is made is also consistent , by using the gray color.

  1. Universal Usability

This application provides Help function so that novice users can use this application and know the basic knowledge of the existing menus. Then for the expert user, this application provided quick functions, so that the user can use the application quickly.

  1. Provide informative feedback

This application provide feedback to the user on the import data interface. If the user import the data with an incorrect format , then the application will provide feedback that is informative and easy to understand .

  1. Designing a dialogue that provides closure

Thisapplicationis designedthatthe user mustperformactionssequentially. Usercan not addthe manifestvariablesbeforeimportingthe data first. Then theusercannotanalyzethe dataif theuserhas not createdrelationships.

  1. Simple error prevention

This applicationalsohas apreventionif the usermade ​​a mistake. Simplepreventionareon the Home interface. If theuserhas not beenestablishedstructuralrelationsor measurement relationsand itdidanalyzethe data,, then the applicationwill displaya warningscreentoperformchecks on thestructuralrelationsor measurement relations thatestablished.

  1. Allow reversal action ( undo )

This applicationallows theusertocancelthe actionandreturntothe previousconditions. Exampleson structuralrelationships interface andmeasurementrelationship interface. When the userestablishrelationships, the applicationprovides acancel buttontoreverse therelationshipthat has beenestablished.

  1. Support the internal locus of control

This applicationalsoallows theusertoset thedesired screen. Allscreensin this applicationcanbe resizedaccording to thewishes of the user.

  1. Reducing the burden of short-term memory

This applicationusesfunctions thatare easy to understandso the userdoes notneed to remember thefunctionsthatare availableinthis application.

ConclusionsandSuggestions

Conclusions

Based on the analysisandapplication outputwithPartial LeastSquare method, then getsomeconclusions, namely :

  1. Structuralmodel ofthenatural resources, human resourcesandeconomicgrowthis

ECONOMIC = 0,27 NR + 0,65 HR

The modelhas aR-square (R2) value 0,51whichmeans thatthis studyhasstructuralmodelsaregood.

  1. Measurementmodel ofthenatural resources, human resourcesandeconomicgrowthis

Fishery = 0,86 NR

Plantation = 0,71 NR

Literacy = 1 HR

GDP = 0,83 ECONOMIC

Expenditures = 0,73 ECONOMIC

The modelhas met theconvergentvalidity, discriminantvalidity, andcompositereliability.

  1. Natrual resourceshave a significant effect toeconomicgrowthat5% significance level. Itis indicatedbyTstatistics value of natural resources is 1,985 that are lessthan Tα/2, dftable is1.960.
  2. Human resourceshave a significant effect toeconomicgrowthat5% significance level. Itis indicatedbyTstatistics value of human resources is 10,435 that are lessthan Tα/2, dftable is1.960.
  3. The PLSfunctionscontained inthis developed apllication are convergentvalidity, discriminantvalidity, composite reliability, andR-squared(R2).

Suggestions

Suggestionsorproposalsforfurther researchare:

  1. Use PLSSEMmethodtoanalyzethe sameproblematanother location.
  2. UsingPartial LeastSquareapplicationsthathavebeen developedtoanalyzetheproblemsin otherfields.

References

Ghozali, Imam. (2011). Structural Equation Modeling Metode Alternatif dengan Partial Least Square (PLS), (edisi 3). Semarang : Badan Penerbit Undip.

Indrasari, Viki. (2011). Analisis Faktor-Faktor yang Mempengaruhi Pertumbuhan Ekonomi Provinsi Jawa Tengah, Semarang.

Mustafa, Zainal E.Q. dan Wijaya, T. (2012). Panduan Teknik Statistik SEM & PLS dengan SPSS AMOS.Yoyakarta : Cahaya Atma Pustaka.

Santoso, Singgih. (2012). Analisis SEM menggunakan AMOS.Jakarta : Penerbit PT Elex Media Komputindo.

Schneiderman, Ben dan Plaisant, Catherine. (2010). Designing The User Interface 5th edition. New York : AddisonWeasley.

Todaro, Michael P. dan Smith, Stephen C. (2006). Pembangunan Ekonomi. Jakarta : Penerbit Erlangga.

Wijanto, Setyo H. (2008). Structural Equation Modeling dengan LISREL 8.8.Yogyakarta : Graha Ilmu.

History of Writer

Melvin Hendri was bornin thecity ofPalembangon June 2,1990. Authoer is graduatedhis bachelor at Bina Nusantara Universityin thefield ofComputer TechnologyandStatisticsin 2013.