Likert Scale Development Petrol Protests, Psychology, College Term Papers.com


		Data Bases
		• Custom Term Papers • Free Term Papers • Free Research Papers • Free Essays • Free Book Reports • Plagiarism?

		Links
		• Top 100 Term Paper Sites • Top 25 Essay Sites • Top 50 Essay Sites • Search 97,000 Papers @ DirectEssays.com • Search 101,000 Papers @ ExampleEssays.com • Search 90,000 Papers @ MegaEssays.com

		Free Essays
		• Term Paper Sites • Chuck III's Free Essays • Free College Essays • TermPaperSites.com • My Term Papers • Get Free Essays • Essay World • Planet Papers

Search Lots of Essays

Back to Subjects - Psychology

Likert Scale Development Petrol Protests

Likert Scale Development Petrol Protests Social psychological attitudes are perhaps one of the most important determinants of society in that it is the individual’s attitude that motivates his or her behaviour. As Rajecki (1990) summed up this conception: ‘...attitude is seen as the cause and behaviour is seen as the effect.’ (Rajecki, 1990, p.4). Given consideration, it is easy to see that this is in fact the case; for example, why do people aspire to an education? The reason for this is that they perceive an education to be a positive thing and therefore as something to be desired. There are many words which are variations of ‘attitude’ such as beliefs, perceptions, convictions, judgements and opinions. Every day we are forming attitudes and acting according to these attitudes and so it becomes clear that attitudes are the factors underpinning everyday life. Allport (1935) defined attitudes as ‘... a mental neural stateof readiness, organised through experience, exerting a directive or dynamic influence upon the individual’s response to all objects and situations with which it is related’ (Rajecki, 1990, p.4). Attitudes are normally measured through the use of scales such as Thurstone’s equal appearing interval scale and Likert’s(1932) summated rating scale. The Likert scale is perhaps one of the most widely used scale in attitude measurement (Dyer, 1995). Unlike the other scales, the Likert scale employs only monotone items which are either favourable or unfavourable. Respondents have to indcate their level of agreement with a list of these monotone items on a scale from one to five. The answer format is normally presented in the following order: strongly agree, agree, undecided, disagree, strongly disagree (See App1, Figure 1 for example). Each of these answers is assigned a score, for example, strongly agree attains a score of 1. One of the advantages of this type of response set is that, not only does it indicate the respondents attitude, also it indicates the strength of that respondent’s attitude towards the construct being measured. This is done by totalling up the responses given by the responsent across all items; this total score can then be evaluated against the total possible score, and as a result, an attitude strength proposed. Generally, in developing a Likert scale, a large number of items which appear to have face validity should be produced. These are then tested on a pilot sample which are the same or similar to the group on which the final attitude scale will be sampled. Once sampled, the items on the scale are then analysed in terms of their validity and reliability and poorly performing items are removed. The items remaining should then all measure the construct of interest in a valid and reliable manner; this final scale is then ready for application to the target population. There are a number of different issues concerning reliability and validity which must be considered. Firstly in the item development, all items should have face and therefore content validity; that is, all the items should appear to be measuring the corresponding domain of interest (Dyer, 1995) The scale should obviously have criterion-related validity, in other words it should be able to differentiate between the two criterion groups being tested. The scale should have internal reliable so that the differences found between the two criterion groups are actually caused by the difference in attitudes towards the construct being investigated. The construct being investigated in this study was people’s attitudes towards the recent fuel-tax protests, when the ‘...fuel protestors- and the beleaguered motoring public...’ articulated their anger at ‘...sky-high fuel duties...’ and ‘...brought Britain to a halt in September.’ (The Economist, Nov. 11th, 2000, p.37). The aim was to develop an attitudinal measure which would successfully differentiate between those people who support the fuel-tax protests and those that are against them. Thus, it was hypothesised that people who obtained a low score on the scale being developed would have previously stated that they were in support of the fuel protests. Conversely, those who obtained a high score would have previously rated themselves as against the recent fuel protests. Quota sampling was employed to select 50 subjects who would participate in this study. 25 of the participants rated themselves as in support of the fuel protests and the remainder rated themselves as against the protests. The participants were preselected in terms of their attitudes towards the protests in order to facilitate later scale analysis of criterion related validity. Of these 50, 23 were male and 27 were female; all were Level 2 Undergraduates from the University of Leeds. Their ages ranged from 18 to 28 years. The participants in this sample were presented with a Likert scale measuring their attitudes towards the recent fuel protests (see App 1). The Likert scale contained 40 items, which were generated at random by the experimenter group, 20 of the items generated were worded favourably while 20 were worded unfavourably. The scale consisted of a variety of items, either of a cognitive, affective and behavioural aspect of the attitude. The items included were drawn from all areas of the content domain. Each of the items had 5 possible responses ranging from Strongly Agree through to Strongly Disagree, the response strongly agree obtained a score of 1 whereas strongly disagree obtained a score of 5. (see App. 1) A cross sectional design was used in this experiment. There were two experimental groups; one, which contained those in favour of the fuel protests, and group two, which contained those in opposition to the fuel protests. 50 percent of the scale items were reversed so to indicate support for the protests, the respondent would have to disagree with the statement. This was employed to lower the incidence of agreement with items regardless of the content. The allocation of the positive and negative (reversed) items onto the scale was randomised so that there was no fixed ordering of the items. The participants were asked whether or not they supported the recent fuel protests and this was recorded on their answer sheet. If the quota for that attitude group was already reached they were thanked but were not used further. The participants utilised were asked to complete the scale and thanked upon completion. Their initials were marked on their scale so that they could be recontacted at a later date. Their agreement scores were then inputted into SPSS for analysis. A scale reliability analysis was run on the results and data obtained for the item means, item standard deviations, inter-item correlations (Pearson’s) and the scale statistics. Item analysis was then carried out in order to reduce the number of items and refine the scale. Firstly, each of the item means were analysed, any item which had a mean outside of the range 2.5 - 3.5 was removed as these items were not efficiently discriminating between the two criterion groups The item means should lie within this range as one group should be scoring highly on each item while the other should obtain low scores on each item. Secondly, the standard deviations of each item was analysed and any that appeared to be different from the majority were removed. Thirdly, the inter-item correlations (Pearson’s correlation coefficients) were analysed, any items which had a relationship of less than 0.5 (Pearson’s r) with the majority of the other items were removed as this implies that they are not measuring the same construct as the majority. Next, the effect that each of the items had on the scale variance was considered. If an item radically effects the variation in the scale then it was deleted. The corrected Item-Total Correlation was then considered; if any item correlated with the total score of the rest of the items by less than 0.5 it was deleted. Lastly, the effect that each item would have on the scale alpha if it was deleted was analysed, if the deletion of any item significantly increased the alpha then it was removed. The scale alpha being Cronbach’s coefficient alpha which represents the mean reliability coefficient that could be obtained from all possible split-half tests of reliability. Items were removed until the number of items remaining totalled 20. The aim of this process was to refine the scale, resulting in a scale that would achieve a higher level of content validity, criterion-related validity and internal reliability than did the initial pilot scale. A reliability analysis was performed on the results using SPSS and the following results were found. As can be seen from Table 1, the scale mean is higher than one would expect, given that there were an even number of participants for and against the protests (Mean = 118.87). A scale mean of closer to 100 would be expected when the total possible score is 200 and the experimental groups should be achieving scores at opposite ends of the score range. Inter-itemCorrelations .4245 -.8298 .8752 Table 2 shows that the overall item mean is 2.97 which was expected, given that 3 was the neutral agreement score. 25 people agreeing and disagreeing should cancel out across all items to an average of 3. As can be seen from the minimum and maximum mean scores there are a number of items which lie outside the required range of 2.5 - 3.5. The mean inter-item correlation for the scale was 0.42 which indicates a low to moderate correlation, obviously, this is not ideal and implies that there are items which don’t correlate very well with the others. This can be seen from the minimum and maximum inter-item correlations of -0.8298 and 0.8752. The scale alpha at this stage was very high, well above the required level (á = 0.9674). Means & S.Ds 9 9,10,15,16,20,23,28,32,36 Correlation Matrix 6 6,8,12,19,26,27 Table 3 shows that twenty items were removed from our item pool; on the basis of the above statistics and their effect on the scale statistics (reasons for item deletion can be seen in more detail in App. 3). The same 50 participants were used in the second scale application as in the application of the initial 40 item scale. Again, the participants were presented with a Likert scale measuring attitudes towards the recent fuel protest. However, the Likert scale consisted of the 20 items judged to be most reliable and valid from the first scale application. The response set again ranged from strongly agree to strongly disagree with a response of strongly agree scored as 1 and strongly disagree as 5. At time two a repeated measures design was used in the second scale application.Similar to time one, two experimental groups were used; one, which contained those in favour of the fuel protests, and group two, which contained those in opposition to the fuel protests. 50 percent of the items were reversed to reduce agreeing regardless of item content. Again ordering was randomised so that there was no obvious positive - negative pattern among the items. The 50 participants from the first application were contacted a week later and asked to complete the second version of the attitude scale. Again their initials were marked on their response sheet so that their responses could be matched to their first set of responses. This was necessary so that the test-retest reliability of the scale could be evaluated. Once they had completed the scale their responses were inputted into SPSS and as before the negatively worded items were recoded so that they matched the positive items. A reliability analysis was performed on the data to identify any items that were performing poorly and they were subsequently removed. Firstly, each of the item means were analysed, any item which had a mean outside of the range 2.5 - 3.5 was removed as these items were not efficiently discriminating between the two criterion groups The item means should lie within this range as one group should be scoring highly on each item while the other should obtain low scores on each item. Secondly, the standard deviations of each item was analysed and any that appeared to be different from the majority were removed. Thirdly, the inter-item correlations (Pearson’s correlation coefficients) were analysed, any items which had a relationship of less than 0.5 (Pearson’s r) with the majority of the other items were removed as this implies that they are not measuring the same construct as the majority. Next, the effect that each of the items had on the scale variance was considered. If an item radically effects the variation in the scale then it was deleted. The corrected Item-Total Correlation was then considered; if any item correlated with the total score of the rest of the items by less than 0.5 it was deleted. Lastly, the effect that each item would have on the Cronbach’s coefficient alpha if it was deleted was analysed, if the deletion of any item significantly increased the alpha then it was removed. A Principal Component Analysis was then performed on the data so that the scale could be analysed in terms of the factors underlying the scale items. A component was deemed to be an underlying factor if it’s Eigenvalue, the amount of variance accounted for by each component or factor, exceeded 1. The presence of more than one factor means that the data was rotated using varimax in order to maximise the relationship between the variables and the factor on which they are loaded. Generally, if an item displays a loading of 0.4 or above on a specific factor, then it is classed as loading on that factor. However, occasionally it is necessary to be more stringent and only accept a loading of 0.5 or greater. This is normally done when there are a large number of ambiguous items which load by 0.4 or above on both factors thus making them difficult to classify. Four items were removed during the reliability analysis meaning that only 16 items were inputted into the factor analysis process (see App 3, Table 2) Table 2 - Reason’s for item removal at time 2 3 The government should give in to the blockades Increases Alpha 6 The government should not give in to bully-boy tactics Mean * 3.5 8 Fuel tax cuts are essential for businesses to survive Mean * 2.5 14 The blockades have achieved good results Mean * 3.5 As can be seen from the table outlining the reasons why four items were removed from the scale before the final analysis and subsequent factor analysis, three of the items were removed because their means fell outside the required range (items 6, 8 and 14) and item 3 was removed because it increased the alpha if deleted. Final analysis of our scale revealed that the mean correlation between items was moderate to high (r = 0.67). This statistic indicates that all scale items appear to be measuring the same construct. Also, the overall item mean across all items is 3.1 which is around the figure expected on a response set containing five levels of agreement. Cronbach’s Alpha for the scale indicated very high internal consistency (á = 0.97). A Pearson’s correlation was carried out on the data from time one and time two and indicated a high level of test-retest reliability (r = 0.92, N = 46, P*0.001). This means the responses remained reasonably constant over the time interval between the two test dates. Further, the criterion related validity was assessed using an independent measures t-test which indicated that the scale possessed a high level of criterion-related validity. In other words, the scale analysis showed that the scale could effectively discriminate between the two criterion groups, namely those who supported the protests and those who did not (t (44) = -11.33, P*0.001). The group who were in support of the protests obtained a mean score of 34.30 whereas the group in opposition to the protests obtained a mean score of 64.91. Principal components analysis was used to identify the underlying factors. This revealed 2 factors, the first had an Eigenvalue of 11.00 and accounted for 68.8 percent of the total variance across all variable scores. Meanwhile, the second factor achieved an Eigenvalue of 1.09 and accounted for 6.83 percent of the total variance. The disruption was for a good cause .832 Fuel protests are fast losing public support .679 I would be without my car to support the fuel protest .726 I am happy someone is taking the initiative .802 If I thought it would help the protestors I would fill up my petrol tank .809 This sort of disruption causes pain and suffering .809 The Government should give in to strong public support .754 Civil talks would be more appropriate than blockades .877 Blockades will change taxation .700 Negotiations are more appropriate than blockades .890 The Government should not be bullied over policies .577 .644 I do not support the methods used by the protestors .594 .681 Blockades are the only way to get government attention .718 .518 The Government is correct in standing firm .606 .638 Blockades did not and will not solve problems .632 .615 Lives were put at risk during the protest .594 .577 Visual inspection of Table 2 reveals that, using the modified rule that a loading of greater than 0.5 indicates loading on one factor and not the other, 7 items load on Component 1 (items 4,9,12,13,17,18,19) while three load on Component 2 (items 1,16,20). However, 6 items remain that load heavily on both factor such that this rule does not apply to them (Items 2,5,7,10,11,15). This means that these items are highly ambiguous and as such, a decision cannot be made with regards to the component onto which they load. Examination of the content of the scale items reveals that the two factors extracted from the data can be labelled as; Factor One: Supporting the cause and Factor Two: Issues concerning blockades. The fact that two underlying factors were extracted during factor analysis indicates, therefore, that the scale is not unidimensional and therefore the scale does not measure only the attitude towards the fuel protests but also towards the issues concerned with the blockading of the fuel refineries The hypothesis being investigated was: those who score low on the scale will have previously rated themselves as in support of the recent fuel protests. This of course means that those who rated themselves as ‘in opposition to the fuel protests’ will score highly. This hypothesis was indeed supported by the results, inspection of the group means and the t-test results showed that the supporting group obtained a low mean score whereas the opposing group obtained a high mean score. This means that the scale does indeed possess criterion-related validity. In addition, it was shown to possess test-retest reliability. However, although Cronbach’s alpha indicated that the scale possessed a very high level of internal consistency, factor analysis revealed that there were two factors underlying the variance between the groups. This ultimately indicates that the scale is not unidimensional and therefore the scale measures more than one content domain. It can be seen that the scores on the scale developed can be explained in terms of two factors: supporting the cause and issues concerning blockades. Even though the two are related the scale does not measure only attitudes towards the fuel protests but also it draws on issues concerning blockades. Thus, replication followed by further item analysis and refining will need to take place before the scale can be successfully validated as a measure of what it claims to be. One reason why there are a number of ambiguous items in terms of the factor onto which they load is that the sample size is insufficient. Kline’s (1981b) study that factors tend to become unclear if the sample size drops below 100 as the larger the sample, the more the standard error of the correlations are reduced. Guildford (1956) stated that 200 subjects is a preferable minimum, whereas Kline (1999) states that 100 is the absolute minimum sample size for a measure involving factor analysis. In addition, for scales with a large number of items, he advised a participants to item ratio of at least 3:1. Kline (1999) also stated that due to the frequency of sex differences appearing, the scale should ideally be run on two separate samples of 100 males and 100 females. According to Kline (1999), the sample population should also be representative of the population for which the scale is intended. This is important in terms of this study because the sample used consisted entirely of students at the University of Leeds. Therefore, perhaps a more representative sample would have positively affected the outcome of this study, in that the construct of interest may hold greater significance for other sections of society. This is due to the fact that few students actually possess cars and therefore the level of fuel tax is perhaps less significant than for those who own and run cars. In addition he suggested that test-retest reliability should be tested after a three-month gap which would give a better indication of this form of reliability. Guildford (1956) also stated that for five-point scales such as the Likert scale, Pearson’s correlation coefficient is not ideally suited because it works more efficiently with scales utilising greater ranges. Cattell (1978) was sceptical about the factors extracted from certain forms of attitude scales. He believed that factors extracted from scales containing ‘duplicated’ items were unreliable because the items were often paraphrased and utilised more than once. He believed this meant that the items would obviously load on a specific factor therefore not giving a true indication of the other factors underlying the scale. Therefore, he suggested that paraphrases of other scale items should almost certainly not be included if a reliable scale was sought. It can be seen that there are many problems concerning scaling and they must be taken into account when developing a scale. In term of this study, further replication is necessary with new items included to refine the scale further. The presence of the ambiguous items poses a problem which Kline (1999) believed could only be resolved by their removal from the scale and the input of new items. Table 1 - Summary Scale and Item Statistics Table 2 - Pearson's r results for Scale TOTAL1 Pearson Correlation 1.000 .922 TOTAL2 Pearson Correlation .922 1.000 Table 3 - Levene’s Test for Equality of Variances Levene's Test for Equality of Variances TOTAL2 Equal variances assumed 2.090 .155 Table 3 - Independent samples t-test for criterion related validity t-test for Equality of Means TOTAL2 Equal variances assumed -11.330 44 .000 Equal variances not assumed -11.330 39.892 .000 1 The blockade was a great idea Increases Alpha 3 I support the recent fuel protest Increases Alpha 6 Taxes are high to lower pollution caused by cars r * 0.6 7 The protestors acted irresponsibly Increases Alpha 8 Petrol prices are high and damaging small businesses r * 0.6 9 The presence of the farmers was justified SD Low 10 If the government gave in to the blockaders it would worry me SD High 12 Vigilante behaviour is unacceptable r * 0.6 14 I would actively support further protests Increases Alpha 15 I do not support the protests SD High 16 Government action during the protest was unacceptable SD Low 19 Petrol prices are damaging the economy r * 0.6 20 I would not support the protests if there was a recurrence SD High 23 Fuel prices are reasonable Mean * 2.5 26 The fuel crisis disrupted my life r * 0.6 27 The government should give in and reduce fuel prices r * 0.6 28 The protest should occur again to reduce fuel prices SD High 32 Protestors should have been arrested Mean * 2.5 36 Petrol prices are too high Mean * 2.5 39 Taxes are high to improve public services Increases Alpha Table 2 - Reason’s for item removal at time 2 3 The government should give in to the blockades r * 0.6 6 The government should not give in to bully-boy tactics Mean * 3.5 8 Fuel tax cuts are essential for businesses to survive Mean * 2.5 14 The blockades have achieved good results Mean * 3.5 I would be happy to have a nuclear waste plant in my back garden 1 2 3 4 5 Strongly Agree Undecided Disagree Strongly Figure 1. - Example attitude statement Bibliography:

Word Count: 4171