AVIATION SAFETY CULTURE MEASUREMENT MODEL FIT VALIDATION OF A SURVEY FOR THE AVIATION MAINTENANCE REPAIR ORGANIZATIONS

It is believed that safety is the major issue for the aviation industry. According to Boeing Study %15 of the fatal accidents are incurred by maintenance sources related. On the other hand, from the last ten years safety management system which firstly introduced by ICAO became popular in the aviation industry. In the beginning, enforcement by ICAO Annex 19 then following it consequtively by EASA’s regulation, at the same time issued many advisory circulars by FAA that SMS entried in force and still other authorities are acting by similar way. However, the most important element of the SMS is based on properly establishment of safety culture in organization. Up to now there are many methods introduced in the literature for measuring safety culture of aviation organizations. Even though an aviation company may select one model as to another one for measuring own safety culture, but they are not sure whether this model run perfectly and reflect true results or not. In this study it is aimed to bring out how the proposed ECAST’s safety culture model which is compliance for the aviation maintenance organizations by validating Structural Equation Modelling.


1.1.SAFETY CULTURE DEFINITION IN AVIATION
Safety culture is a term that nearly everyone uses but few can agree upon its precise meaning or how it can be measured. The social as a first it comprises the beliefs, attitudes and values often unspoken of an organization's membership regarding the pursuit of safety and the second is more concrete and embraces the structures, practices, controls and policies that organizations possesses and employs to achieve greater safety (J. Reason, A.Hobbs 2003). A Safety Culture refers to the extent to which every individual and every group of the organization is aware of the risks and unknown hazards induced by its activities; is continuously behaving so as to preserve and enhance safety; is willing and able to adapt itself when facing safety issues; is willing to communicate safety issues; and consistently evaluates safety related behaviour (P. Montijn, B.NLR, 2009)

SAFETY CULTURE REQUIREMENTS FOR AVIATION
When it comes to review the Space Shuttle accident's executive report in 2003, it will be seen that the organizational causes of this accident are rooted in the Space Shuttle Program's history and culture. Additionally, cultural traits and organizational practices detrimental to safety were allowed to develop (Patrick Hudson, 2007). it may be cited as an examples (IAEA, 2014) because of national culture which is sub-components of the safety culture caused five serious aviation accidents that are Tenerif -Runway Incursion -Canary Island, Sprain -1977 (583 dead), Avianca 052 -Crash -New York -1990 (73 dead), Korean Air 801 -Crash -Guam -1997 (228 dead), The Überlingen mid-air collision -Switzerland -2002 (71 dead), Asiana 214 -Crash -San Francisco -2013 (3 dead). These samples can be risen however, nowadays it is considered that safety culture still is the most important problem in order to set up safety management system (Ender Gerede, 2012). Since safety culture is almost the root cause of the errors which are occurred by based on general system (James Reason, 2013, 81).

1.2.AVIATION SAFETY CULTURE MODELS
There are many kind of model for measurement and assessment of the Aviation Safety Culture in the literature. Reason

ECAST (EUROPEAN COMMERCIAL AVIATION SAFETY TEAM) SAFETY CULTURE MODEL
As Safety Culture is still an emerging issue and since the introduction of the ICAO and EASA Safety Management requirements will expedite Safety Culture activity across Europe. (Piers, Montijn & Balk, NLR 2009, Safety Culture Frame Work for the ECAST-Working Group).
Nevertheless, a strong Safety Culture is generally considered as a vital condition to a wellfunctioning of SMS. For this reason, the SMS Working Group of ECAST has been tasked to propose Safety Culture reference material.

ECAST'S PROPOSED SAFETY CULTURE FRAMEWORK
From the review of the main existing and emerging Safety Culture frameworks in aviation and beyond, we know that Safety Culture is a multi-dimensional construct. To capture the common . When we closely looked in these six characteristics they will be represented by some indicators which are shown below:   Another outstanding safety maturity model had been developed by also Fleming (Fleming Mark, 2000, 2000. The safety culture maturity of an organisation consists of ten elements, which are described as management commitment and visibility, communication, productivity versus safety, learning organisation, safety resources, participation, shared perceptions about safety, trust, industrial relations and job satisfaction and training. This model has been adapted from the safety culture components listed by the HSE in HSG487 (HSE, 1999).

RESEARCH METHODOLOGY
It is aimed by this application to determine whether or not ECAST's proposed Safety Culture Model is suitable to measure the safety culture levels of certifying staff, mechanics, noncertifying staff and whole group in a maintenance repair organization via statically model validation analysis.

RESEARCH TECHNIQUE AND CONSTRAINTS
A survey method developed in six dimensions and having a Likert type scale of 5 by EASA Safety Management Group by using Reason method and taking ICAO criteria into consideration. The research is conducted in Maintenance Repair Organization located at Turkey by using survey method on Certifying Staff, Mechanics and Non Certifying Staff, which are representing the majority of company production staff.

DATA COLLECTION
The questionnaire is tested before the using in the field via pilot study on each group's representative sampling. At the two locations, each group who are certifying staff, mechanics and non-certifying staff are invited to the classroom in different days. First of all, they are informed respectively about to survey. Then, the representative of each group who are 25 staff fill in questionnaires since it has been translated from English to Turkish language. Thereby, 52 survey questions are checked whether they are understood by each group correctly or not. On the other hand, the purpose of time study is to determine average response time of one question because of the survey is containing 52 questions. Questionnaires are distributed to all attendances and collected back after they are filled by company safety officers. Filling the questionnaire process is conducted under the supervision of safety officers. Therefore, misunderstood questions, response time etc. have been corrected. After then, in accordance with sampling model, representative of each group member respectively are invited to the classroom for filling the questionnaire in different days at two locations (facility-1 and facility-2). Finally, the concerned survey has been conducted within two months.

DESIGNING SURVEY QUESTIONNAIRE
Questionnaire form which designed by EASA Safety Steering Group, scaled by 6 dimensions and consists in 52 questions for using aviation safety culture levels measurement. The responses are addressed on a 5 point Likert type scale ranging from 1 to 5 (1= Totally Agree, 5= totally disagree). This questionnaire used by translation of this scale.

ANALYSIS METHODOLOGY
The data analysed by means of SPSS and AMOS statistic software. The collected data processed by outlier analysis, homogeneity analysis for each group then exploratory factor analysis, reliability analysis are performed for each dimension of the scale by using SPSS. Thereby, uncorrelated and insignificant questions were extracted from its concerned dimension. To continue with that by means of structural equation model of the AMOS software, first order conformity factor analysis performed for the model goodness fit and validation of scale's dimensions. Finally, by the path analysis method the concerned scale was tested for each group whether they are significant to explain and aligned in response to the proposed model or not.

3.1.THE PROPOSED RESEARCH MODEL
As It can be seen that Safety Culture Model scale' consist in 6 dimensions. The model has been shown as below:

Figure 2: ECAST's Safety Culture Measurement Model in Aviation
In this study, safety culture maturity levels measured not only certifying staff but also mechanics and non-certifying group. Thereby, total company safety culture is depend on the each group safety culture which are presented as following figure.

NORMALITY TESTS
According to the Kolgmogrov-Smirnov Normality Test which run for whole group, since variable of the safety culture Sig.value >0.005 then H0 hypothesis accepted and data are distributed normally.

OUTLIERS ANALYSIS
According to the certifying staff's outlier analysing results, the number of data 7,45,61 and 107th accepted as outliers then they were omitted from the data list.

HOMOGENEITY TEST
According to the results of homogeneity test for the certifying staff, Sig. values >0.05 thereby homogeneity is validated.

FACTOR ANALYSIS AND RELIABILITY ANALYSIS
For each dimension factor analysis performed at the first step and then reliability analysis performed at the second step after uncorrelated and insignificant questions extracted from the its dimension. In order to examine reliability of structure which formed by each latent variable reliability of structure, the proportion total variance explained and Cronbach's alpha reliability criteria are calculated. Nunnalyy argue that each of structure Cronbah's alpha value must be greater than 0.70 in social sciences for reliability of measurement tool (Nunnaly, J.C. and Bernstein, I.H., 1994). Some authors suggest Measure of Sampling Adequacy which developed by Kaiser to apply the factor analysis model successfully (Rencher A.C, 1998). Hair and Et are considered to be sufficient coefficient value for reliability of structure must be greater than 0.70 and the proportion of total explained variance values must be greater than 0,50 for the latent variables (Hair, 1998 (Tavşancıl, 2005, 399). Tebachnic and Fidel assert that decision of the size of the factor loadings which are acceptable should be given by researchers (Tebachnic and Fidel, 2001). Therefore, the relevant scale has been ready for the analysis of the structural equation modelling.

DESCRIPTIVE STATISTICS AND DEMOGRAPHICS
Descriptive statistics for each group are as follows: Certifying Staff: All certifying staff (110) gender are consist of "male" and selected 78% of the group from facility-1. The department distribution of certifying staff is base maintenance 80% and component maintenance 20%.
Mechanics: There are only one woman and 91 men in mechanics group. Mechanics are selected from facility-1 as 53% and facility-2 as 47%. The department distribution of mechanics is base maintenance 71% and component maintenance 29%.
Non-Certifying Staff: There are 6 women and 214 men in non-certifying staff group. Noncertifying staffs are selected from facility-1 as 33% and facility-2 as 67%. The department distribution of non-certifying staff is base maintenance 67% and component maintenance 33%.
Descriptive statistics which are indicated mean values and standard deviation of each dimension and safety culture levels of Certifying Staff, Mechanics and Non Certifying Staff. As it can be seen from the table while the CS and Non-CS means of safety culture level is equal which is 3.2 therefore the safety culture level of Mechanics is slightly higher than others that is

CONFIRMATORY FACTOR ANALYSIS (CFA) AND STRUCTURAL EQUATION MODEL (SEM)
Confirmatory factor analysis (CFA) is an extension of factor analysis which tests whether a set of items defines a construct (Schumacker & Lomax, 1996). A Confirmatory Factor Analysis was conducted to validate the measurement model of each latent construct. The CFA evaluates the construct validity of the proposed model to determine whether it is intended to measure what it is supposed to measure (Kline, 1998). The CFA was performed using AMOS 22 software. Goodness of fit indices are used to determine how well the model fit the collected data. A single fit index is not enough to support the fitness of the model to a given data set (Vandenberg & Scarpello, 1990 a ratio of chi-square to degrees of freedom of between two and three represents an acceptable fit. The second and third fit indices are the comparative fit index (CFI) and the Tucker-Lewis index (TLI). The CFI compares the hypothesized model with a null model and considered to be reasonably robust against violation of assumption. A value above 0.95 is considered good whereas a value between 0.90 and 0.95 is acceptable. The TLI is used to compare a single model or alternative models to a null model and is less sensitive to sample size. A value of more than 0.95 indicates a good fit while a value between 0.90 and 0.95 is acceptable. In addition, a value of less than 0.90 requires a restructure of the model. The fourth and fifth indices are goodness of fit index (GFI) and adjusted goodness of fit index (AGFI). Theses indices are developed by Joreskog and Sörbom. GFI has been developed alternatively to the chi-square in order to evaluate fit of model independently from sample size. Indices of GFI and AGFI changes between 0 and 1. AGFI is an arranged derivation of the GVI to estimate parameters numbers. A value above or equal 0.90 is considered good A value above or equal 0.90 is considered good and a value equal 1 is considered perfect (

3.1.6.2.BEHAVIOUR
The dimension of the behaviour, AMOS trimmed model solution and model fit values are resulted in acceptable as seen below scheme and in table 6 for Certifying Staff.

3.1.6.3.AWARENESS
The dimension of the awareness, AMOS trimmed model solution and model fit values are resulted in as seen below for Certifying Staff.

3.1.6.4.ADAPTATION
The dimension of adaptation AMOS trimmed model solution and model fit values are resulted in as seen below for Certifying Staff.

3.1.6.5.INFORMATION
The dimension of the information, AMOS trimmed model solution and good of fitness values are resulted in as seen below for Certifying Staff.

3.1.6.6.JUSTNESS
The dimension of the information, AMOS trimmed model solution and good of fitness values are resulted in as seen below for Certifying Staff.

3.2.THE FIRST ORDER CONFIRMATORY FACTOR ANALYSIS AND STRUCTURAL EQUATION MODEL
When we performed the first order confirmatory factor analysis for Certifying Staff in all dimensions by using SEM we would be got following modified model. The concerned SEM l regression weights and estimated values are acceptable levels and seen as follows:

CERTIFYING STAFF , MECHANICS, NON-CERTIFYING STAFF PATH ANALYSIS
Confirmatory Factor Analysis which explained above not only performed Certifying staff but also performed for the Mechanics and Non Certifying Staff. Therefore, when it comes to path analysis for each group, as specified in below, by using the latent variables the related structural equation model developed and model fit values calculated for each group.

Figure 11: Certifying Staff ECAST Model Path Analysis
The concerned SEM 2 regression weights and estimated values are acceptable levels and seen as follows          To conclude with that, Westrum's safety culture measurement model can use in maintenance repair organizations. There are more than 500 maintenance repair organizations which may benefit from this model in the EU and Turkey. Finally, this study is addressing safety culture measurement model fitness in the maintenance repair organizations not the others aviation operations therefore it must be validated before using others type of aviation operations.