Predictors of Work Injuries: A Quantitative Exploration of Level of English Proficiency as a Predictor of Work Injuries in the Construction Industry

ABSTRACT

Both labor shortages and language barriers are present in some areas of the construction industry. These issues translate into higher than average accident and death rates seen among workers of Hispanic origin and the possible confounding effect of a language barrier. In order to address these concerns, an instrument was developed to identify the level of English proficiency. This was compared to injuries sustained on the job in the last 30 days. Participants were ranked on their level of English proficiency, and then divided into two groups. The middle group of predominately bilingual individuals was removed in order to maximize any differences between the two groups. The two groups were then compared based on the different independent variables identified in the research questions, as well as other possible relationships of interest as identified in the auxiliary findings. The overall findings of this research indicate no significant differences in work injuries when based on level of English proficiency. The study revealed the greater the English proficiency; the greater the likelihood a person would experience an injury. This contradicts some findings in the literature identifying Spanish speaking individuals as having a higher accident rate than their more English speaking counterparts.

Keywords:

• safety
• injury
• English proficiency

Introduction

Introduction and Background

In the findings of Fullerton (1999) and Cattan (1993), the Hispanic percentage of the labor force will continue to grow well into the year 2008. With Spanish speakers in the work force, it may pose a problem in the safety sector. In an article written by Hopkins (2003), the year 2000 saw about 815 Hispanic deaths. Those deaths occurred mostly in the construction trades, often due to language barriers, or gaps, impeding communication between supervisors and immigrant workers. According to Flory (2001), “The United States has become an ever-growing multicultural society, especially in large metropolitan areas. The U.S. Census Bureau projected 12 percent of the country's population would be of Hispanic origin by the end of the year 2000. A high number will land jobs in the construction industry, where many of the fatalities and injuries are occurring” (p. 37).

In Table 1, data from Hopkins (2003), adapted from the Bureau of Labor Statistics, show the disparity between race and workplace fatalities. Workplace fatalities decreased for every race on the list from 1992–2000, except for Hispanics, which increased 53 percent.

Table 1. Change in workpalce fatalities, by race, from 1992 to 2000

Race	Percent change
Whites	−10%
Native Americans	−8%
Blacks	−7%
Asians	−4%
Hispanics	53%

According to Lin and Mills (2001), most of the research being done in the occupational health and safety field demonstrates inadequate or non-existing occupational health and safety programs as the primary cause for high rates of injury on the job. Lin and Mills (2001) state “the provision of safety equipment alone does not improve construction site safety, there also needs to be a corporate culture that encourages its use” (p. 132).

Nishgaki (1994) believes several main causes of occupational health and safety failure exist. Those failures include: “inadequate safety education, inadequate instruction, poor housekeeping and ‘willful transgression’” (Lin & Mills, 2001, p. 132). Inadequate safety education and inadequate instruction are two factors directly related to safety training, and can be compounded by a language barrier.

Another way to deal with cultural issues, either organizational or cross-cultural, is to have an individual who identifies with the culture involved provide the training. If the individuals being trained identify with the trainer, their behavior is more likely to change (Kurtz et al., 1997, as discussed in Harvey et al., 2001, p. 620). Misunderstandings between cultures must be dealt with in order for accidents to decline. For safety to be improved, empathy and understanding between the cultures involved must be present (Dake, 1992; Vlek, 1995; Sauer, 1996, as discussed in Harvey et al., 2001, p. 620).

Safety training programs must adapt to the diverse employees of today's workforce. Due to labor force expansion, “the Hispanic labor force has grown by 38 percent” (Martel and Kelter, 2000, p. 13). Although part of this expansion is attributed to population growth, the authors stress “employment grew faster for minority workers than for whites in 1999. The number of employed persons increased by about 2.0 percent for blacks, to 15.2 million; 6.0 percent for Hispanics, to 14.0 million; and 1.2 percent for whites, to 112.7 million” (p. 15). In a projection of the labor force for 2008, Fullerton (1999) states “the 2008 labor force is expected to have a greater proportion of women and Hispanics than the 1998 labor force” (p. 32).

Two main schools of thought exist relating to safety programs: (1) Behavior-based safety programs and (2) incentive based safety programs. According to Geller (1998, as discussed in Miozza & Wyld, 2002, p. 23), behavior-based safety programs work by promoting safe behavior and attempting to decrease unsafe or risky behaviors. These types of programs are “generally embraced by management, but frowned upon by labor. Labor feels that when management engages in a behavior-based safety program, they are, in essence, shifting responsibility for safety to the employees” (Miozza & Wyld, 2002, p. 23).

As Lawrence and Flanders (2000) describe, OSHA classifies incentive-based safety programs in two ways; traditional and non-traditional. Traditional programs reward employees when the company goes a certain period of time with no lost-workday accidents. In contrast, non-traditional programs give rewards for “attending safety meetings, identifying hazards, [and] making suggestions” (p. 30).

According to Fullerton (1999) in a study of projected labor in 2008, “minority groups that have grown the fastest in the past, Asian and others and Hispanics, are projected to continue to grow much faster than white non-Hispanics” (p. 20). Fullerton continues, “although growth of these groups is expected to slow from 1998–2008, the projected growth rates for these groups are nevertheless much faster than for other groups” (1999, p. 22). According to Nixon and Dawson (2002), “by the end of this decade, Hispanic Americans are expected to be the largest minority in US industry and commerce” (p. 189).

With this ever-increasing number of Hispanic workers entering the labor force, the workforce in general is becoming more culturally diverse. Because of this issue, how to effectively train workers, especially relating to safety, has become a greater concern. “Prior policies attempted to treat everyone the same – but everyone is not alike. Personnel policies and management techniques must change to deal with the diverse workforce” (Allen, 1992, as cited in Nixon & Dawson, 2002, p. 185).

Statement of the Problem

As evidenced by a review of the literature, the key issue centers on the higher than average accident rate among Hispanic workers and the possibility of a language barrier contributing to the problem. Language barriers have been documented in other areas, such as agriculture and health care (mainly relating to quality of care received in hospitals), but literature is difficult to find relating to predictors of work injuries in the construction industry. No articles were encountered attempting to compare two groups, based on their level of English proficiency, on predictor factors (i.e., job tenure, impulsivity, safety knowledge, physical hazards, etc. (Frone, 1998)). This study examines if a difference in level of English proficiency can be a predictor of work injuries and accidents.

Significance of the Study

This study is significant because it attempts to determine if a difference in work injury patterns between employees of different English proficiency levels is present. If one group is more susceptible to injuries than another, further research can be done in an attempt to isolate those factors. By so doing, greater emphasis can be placed on these areas to reduce factors contributing to injuries. “Although many studies have examined the predictors of physical and mental health among employed individuals, much less research has focused on behavior health outcomes such as work accidents and injuries” (Jex & Beehr, 1991).

Another area of significance is identified through the body diagram. An indication of where construction workers, regardless of level of English proficiency, are being injured will be given. Through this, precautions can be taken to prevent these types of injuries from occurring. Additional relationships between behavior and work injuries may also provide areas for further research.

Limitations and Assumptions

The sample selection is from one national construction company. Participation in the study was on a volunteer basis. The relationships found in the study might be affected due to the nature of self reported data; participants may not have answered all questions truthfully. In addition, this study focuses on level of English proficiency among Hispanic workers, since the literature identifies this population as being most affected by the possible existence of a language barrier.

This study had very limited funding that served to cover suppliers and lunch for the study participants. With additional funding, the scope of the project could be increased to include more companies, from various geographical regions, which would improve the ability to generalize to larger populations.

Delimitations

The study was delimited to one national construction company. The study was delimited to two construction sites.

Methods

Research Design and Methods

Frone (1998) studied the predictors of work injury among adolescents and compared much of the data to adult employment literature. Because of this comparison, many of the instruments Frone used were adapted for use in this study. This study consisted of several predictor variables believed to influence job injuries. Based on prior research, Frone (1998) chose the following general predictor categories: demographics, personality, health, substance abuse, and employment. Each of those categories was broken up into several sub categories. Combinations of those categories were used in this study in an attempt to determine the predictors of work injuries among adults, and if there was a difference between levels of English proficiency when it came to those predictors.

Demographics

In addition to age and gender, both addressed in the study by Frone (1998), ethnicity and predominant language were examined. Although age was very important to the adolescent study, the predominant language and level of English proficiency were the most important in this study. Little research has been done on accident and injury predictors (Frone, 1998), and no research regarding predominant language as a predictor of work injuries was identified. This study examined if a difference in accident predictors based on predominant language is evident.

Personality

Because of impulsive behavior, some employees “may rush to complete a task without adequate consideration of following safe operating procedures, resulting in increased risk of injury” (Frone, 1998, p. 567). Although prior research has not “directly examined this relation, indirect evidence for it exists among adults” (Frone, 1998, p. 567). Frone believed impulsivity would be positively related to work injuries. The same hypothesis was used in this study. Additionally, for the purpose of this study, the hypothesis indicated the possibility of a significant difference in the predictors of work injuries between varying levels of English proficiency among employees.

Employment Characteristics

The variables in this category studied by Frone were extensive. For the purpose of this study the following predictor variables were examined: job tenure, number of hours worked, and physical hazards. An additional predictor variable attempting to identify level of English proficiency was also added.

Relating to job tenure, Iverson and Erwin (1997) found no correlation to work injuries. Frone, owing to the inconclusive nature of the study by Iverson and Erwin, along with several other studies, presented no hypothesis for the direction of the relationship. Iverson and Erwin believed the longer a person worked a job, the more responsibility they had. Consequently, they would be assigned higher risk jobs because they knew what they were doing.

For work experience, the same inconsistencies were found as in job tenure. According to Frone, “because of potential conceptual inconsistencies and lack of empirical research regarding work hours and injuries…no hypothesis is offered regarding its direction” (p. 567). The same rationale was used in this study; no direction was given for the relationship between work hours and injuries.

According to Frone, prior research carried out by Harrell (1990), Macdonald (1995), and Savery and Wodden (1994). “supports a positive relationship between physical hazards and work injuries” (p. 567). On the issue of workload, Frone, despite the lack of prior research, believed a positive relationship between workload and work injuries would occur.

One additional variable was added from a study by Probst and Brubaker (2001); safety knowledge. Although this study focused on the effects of job insecurity on safety outcomes and compliance, the questions used by the authors were good indicators of safety knowledge. In addition to the three questions taken from the Probst and Brubaker study, several additional questions were developed for this study and added to the already developed questions. The added questions sought to identify company specific safety knowledge.

Participants and Site

Employees of a national construction company, in addition to employed subcontractors of the same company, were used in this study. The data were collected in the spring of 2004. The total population of construction workers surveyed was about 200 out of a possible 700. The total population was sorted for analysis based on level of English proficiency. The study took place at the construction site in order to limit the impact on the work in progress.

Participants were selected on a volunteer basis. Twice a week, over a period of about one month, groups of approximately 25 employees participated in the study. There were 191 total participants in the study. Of those participating, 52 were classified as more Spanish speaking and 102 were classified as more English speaking. The remaining 37 individuals were classified as bilingual, and removed from many of the analyses to maximize group differences. The instrument was administered during the employee's scheduled lunch break. The company provided lunch for participating employees on the days the study took place. The study was administered by the researcher and carried out by an employee of the company who was bilingual.

Data Collection, Instrument, and Procedures

The data was gathered using several Likert scales, the body diagram, the faces of pain scale (developed by Wong and Baker, 1988), and questions relating to general demographic information. The faces of pain scale and the body diagram were both used in several agricultural studies targeting migrant agricultural workers (Faucett et al., 2001). Little modification to these instruments took place due to their previous use among Spanish speaking populations, and their pictorial nature. The overall injury score generated from the body diagram provided “an overall evaluation of symptoms that can be readily compared among populations” (Faucett et al., 2001). The Likert scales, adapted from Frone (1998), were originally used in a study among employed adolescents looking at predictors of work injuries. Although none of the instruments were copyrighted, written permission was obtained from Frone and Faucett. There were no restrictions on using the faces of pain scale developed by Wong and Baker in a non-copyrighted instrument.

The Likert scale portion of the instrument, previously developed by Frone, was modified in order to address methodological issues when administered to Hispanic respondents. The scale anchors from the original Likert and response scales were modified and translated into Spanish. The reasoning is “Likert type anchors typically have no meaningful equivalent in Spanish” (Lange, 2002, p. 412). In a study by Bernal, Wooley, and Schensul (1997), they discovered anchors like “I don't feel sure, I feel a little sure, I feel more or less sure, and I feel very sure” were much more meaningful. When using Likert scales among Hispanic Americans, Lange (2002) believes values must be reassigned so larger numbers correspond to the positive responses.

When the instrument was administered, steps were taken in order to assure participants unfamiliar with Likert scales were able to answer appropriately. A brief presentation with examples and instructions was given in English and Spanish to assure all participants understood how to correctly fill out the instrument. An example, such as “I like to eat candy” might be given, with the responses of agree or disagree available for answers. Once answered, the participant might be asked if they agree a little or a lot. If an example is needed, questions from the survey will not be used. Due to low literacy rates among some populations involved in the study, the researcher, or an interpreter, was available to assist participants in the completion of the instrument. Although there was a potential to influence responses, the risk was taken in order to attain more meaningful responses overall (Lange, 2002).

In addition to the quantitative information obtained from the instrument, one qualitative question was asked. This question was open-ended and focused on other aspects of workers' jobs they believed could cause injury. This was done to try and capture other aspects of how working for the participating national construction company could affect their health.

Instrument Development

The final methodological issue to be addressed is the development of the instrument. The instrument was translated into Spanish for participants with limited English proficiency. To begin with, the instrument was modified to be in the present tense, in the active voice, and at a sufficiently low reading level for all populations involved. Meaning was stressed more than attempting to keep the exact wording.

Once translated into Spanish, the instrument was back-translated into English by a translator not associated with the initial translation. Constructs from the original English version were compared to the constructs emerging from the back-translation and corrections were made as needed to assure congruency between the Spanish and English version of the instrument (Lange, 2002).

The instrument was pilot tested among Spanish and English speaking subgroups to make sure the instrument was clear and easy to understand. During this test, length of the instrument was noted and some issues relating to appropriate completion were found. For one, the body diagram was placed as the first page of the instrument so the oral instructions were still fresh in the participant's minds when they completed the instrument. Another change resulting from the pilot test resulted in the final version of the instrument not being printed on the front and back of each page because some sections on the backs of pages were missed in the pilot test. Feedback from the participants in the pilot test was obtained to ensure perceived meaning matched the intended message (Marín & Marín, 1991). Several minor revisions of Spanish words, resulting from the pilot test, contributed to a clearer understanding of the research instrument.

Through the use of funding given to the researcher by the construction company, the biggest change in the instrument was the use of Colorado State University's testing center in developing the final instrument. Through the use of a computer software program, Design Expert, the previous instrument was adapted into a form easily scanned electronically upon completion. This was of great help in the data entry phase. The new instrument was also much easier to follow.

Data Analysis

Once all the data were gathered, statistics were run looking for significant differences between levels of English proficiency and work injuries. This was done to identify if level of English proficiency contributed to the higher than average accident rates among Hispanic workers in construction. Additionally, each research question was tested by correlation statistics in order to find if positive relationships existed between the predictor variables and work injuries. Results of zero order correlations were presented in a correlation matrix. Effect size was reported for statistically significant relationships to further emphasize statistical significance.

The predictor variables were analyzed one at a time using linear regression. The last step in the data analysis was multiple regression. All independent variables were introduced into the equation to understand the interaction on the dependent variable. It was important to understand how the variables interacted independently and together when comparing predictors of work injuries to level of English proficiency. The research questions are as follows:

1. Based on the body diagram, where are construction workers most likely to be injured?

2. Is there a difference between level of English proficiency and safety knowledge?

3. Is there a difference between level of English proficiency and impulsivity?

4. Is there a difference between level of English proficiency and job tenure?

5. Is there a difference between level of English proficiency and job hazards?

6. Is there a difference between level of English proficiency and work injuries?

7. Is there an association between impulsivity and work injuries?

8. Is there an association between job tenure and work injuries?

9. Is there an association between safety knowledge and work injuries?

10. Is there an association between physical hazards and work injuries?

11. Is there an association between level of English proficiency and work injuries?

12. Is there a combination of impulsivity, job tenure, physical hazards, and safety knowledge that predicts work injuries better than any one predictor variable in isolation?

13. Is there a combination of impulsivity, job tenure, physical hazards, and safety knowledge, based on limited English proficient workers, that predicts work injuries better than any one predictor variable in isolation?

14. Is there a combination of impulsivity, job tenure, physical hazards, and safety knowledge, based on workers that are not limited English proficient, which predicts work injuries better than any one predictor variable in isolation?

Analysis of the Data

Overview

The main variables identified by the research questions are level of English proficiency, safety knowledge, impulsivity, job tenure, overall injury score, and job hazards. Only the data provided in response to the actual research questions will be discussed here. Analysis involving questions introduced while analyzing the data, or other areas of interest allowed by further data obtained by the research instrument, will be discussed in the auxiliary findings section.

Issues arising from data analysis were solved as they surfaced. In order to account for some incomplete survey instruments, it was decided to use the mean function in SPSS, which enabled the researcher to use survey instruments where some questions were left blank. In the case of the age and years questions, written responses were taken over the filled in bubbles. Mistakes were deemed less likely when writing the person's age versus filling in the corresponding bubbles. On the questions to identify level of English proficiency, where two bubbles were filled in for the same question, the more English proficient of the two responses was used. In this comparative study we used the guidelines presented by Gliner and Morgan (2000) to determine the minimal number needed in each cell to determine the maximum statistical power.

For level of education, when a range was identified by methods other than filling in the corresponding bubbles, for example, 1 through 6, the highest part of the range was used. If the range was identified in the bubbles as 7 through 9, and filled in as 79, the higher of the two grades was used for data analyses. In the instances where a face was filled in instead of a bubble for the faces of pain scale, the corresponding number was entered for data analyses. If two bubbles were filled in for the faces of pain question, the lower of the two scores was used. If any responses were too light for the computer to detect, or were filled in with crayon instead of pencil, the corresponding information was entered manually. If yes and no were identified for the company safety questions, the area was left blank for purposes of data analyses.

Examination of the Research Questions

Research question one involves descriptive statistics of the data collected to address frequency and location of work injuries. Additional descriptive statistics obtained from the research instrument are also presented here. Relating to gender of the participants, of the 191 completed surveys, 160 identified themselves as male, and 10 identified themselves as female. The remaining 21 surveys were not marked. Of all participants, 52 were more Spanish speaking and 102 were more English speaking. The remaining 37 were classified as bilingual, and removed from many of the analyses to maximize the group differences. Out of the 191 surveys, several were not completed in their entirety, resulting in varying numbers of useable surveys depending on the variable being used.

Based on research question one for this study, the ten most likely places for construction workers to be injured, solely regarding location, are displayed in Table 2. The ten most likely places to be injured based on type of pain (i.e., sharp, sore and heavy, or numb and tingling) are displayed in Table 3. This is strictly a descriptive look at the location of work injuries. Nothing was found in the literature to indicate different body areas as having higher frequencies of injuries. Lower back, shoulders, foot and calf, and knee were reported as the most common areas to receive an injury. This could give a good starting point for preventive measures to reduce work injuries. Possible solutions could be changes to policies regarding personal safety equipment such as back braces or support for the lower back, or looking to see if employees' footwear is partly to blame for the injuries. In any case, proactive ergonomics or personal safety equipment could be a possible solution.

Table 2. Rank of injury frequencies, location

Rank	Location	Frequency
1	Lower Back, both Sides	128
2	Back Right Shoulder	48
3	Back Left Shoulder	42
	Back Left Foot & Calf	42
4	Back Right Foot & Calf	40
5	Front Right Knee	39
6	Back Middle Spine	38
7	Back Neck	35
8	Front Left Knee	34
9	Front Right Shoulder	23
	Back Right Hip & Buttock	23
10	Front Right Foot & Shin	22
	Back Left Hip & Buttock	22
	Back Left Knee	22

Table 3. Rank of injury frequencies, type of pain

Rank	Location	Type of pain	Frequency
1	Lower Back, both Sides	Sore, Heavy, Tiring	91
2	Back Right Shoulder	Sore, Heavy, Tiring	48
3	Back Left Shoulder	Sore, Heavy, Tiring	42
	Back Left Foot & Calf	Sore, Heavy, Tiring	42
4	Back Right Foot & Calf	Sore, Heavy, Tiring	40
5	Front Right Knee	Sore, Heavy, Tiring	39
6	Back Middle Spine	Sore, Heavy, Tiring	38
7	Lower Back, both Sides	Sharp	37
8	Back Neck	Sore, Heavy, Tiring	35
9	Front Left Knee	Sore, Heavy, Tiring	34
10	Front Right Shoulder	Sore, Heavy, Tiring	23
	Back Right Hip & Buttock	Sore, Heavy, Tiring	23

Table 4 lists many of the descriptive statistics for the study, and Figure 1 identifies the area and frequency of injuries experienced in the last 30 days. An underlined number (blue in the instrument) indicates sore, heavy or tiring pain. A bold number (red in the instrument) indicates sharp pain. An italicized number (yellow in the instrument), indicates numb or tingling pain.

Figure 1 Frequency, location and injury type (underlined = sore, heavy or tiring, bold = sharp, italicized = numbness or tingling).

Table 4a. Descriptive statistics

	Age	Level of schooling	Years working const.	Recode of hours worked last day	Faces of pain scale	Sum of level of English prof.
Mean	34.09	12.00	11.36	9.03	1.50	14.04
Median	32.00	12.00	9.00	8.00	1.00	17.00
Std. deviation	11.19	2.98	9.19	1.49	1.17	6.28
Skewness	0.88	−0.83	1.41	1.05	0.98	−0.43
Std. error of skewness	0.18	0.18	0.18	0.18	0.18	0.18
Minimum	18.00	1.00	1.00	6.00	0.00	0.00
Maximum	66.00	19.00	46.00	13.00	6.00	20.00
N = Valid	188.00	182.00	182.00	190.00	187.00	191.00
= Missing	3.00	9.00	9.00	1.00	4.00	0.00

Table 4b. Descriptive statistics

	Mean impulse score	Mean safety knowledge	Mean physical hazard	Mean of company safety questions	Sum of overall injury score
Mean	2.52	5.14	3.18	0.16	6.35
Median	2.50	5.50	3.14	0.20	5.00
Std. deviation	0.95	1.20	0.73	0.19	6.31
Skewness	0.44	−2.19	0.10	1.33	2.50
Std. error of skewness	0.18	0.18	0.18	0.18	0.18
Minimum	1.00	1.00	1.57	0.00	0.00
Maximum	5.33	6.00	5.00	0.80	47.00
N = Valid	190.00	189.00	181.00	186.00	188.00
= Missing	1.00	2.00	10.00	5.00	3.00

Since the two variables in research question three were approximately normally distributed, an independent samples t-test was run to identify if differences between level of English proficiency and impulsivity would arise. The resulting t-test was not significant, indicating no difference in impulsivity based on level of English proficiency. It would appear both groups are similar in their level of impulsivity.

Research questions four and six (research question five will be discussed later) sought to find differences in job tenure, and work injuries based on level of English proficiency. Since at least one of each variable was skewed, the Mann-Whitney U test was used. The results indicate no significant difference between job tenure and overall injuries, based on level of English proficiency. Both would have been significant at the one-tailed level with significance values of .074 for job tenure and .095 for overall injury score. The literature would have supported a one-tailed test for significance, but in the opposite direction than the one found here, with more Spanish, not more English, reporting the higher injury scores. The resulting r squared value is .028, with a small effect size of .167. This finding would indicate no difference in injuries based on level of English proficiency.

Regarding the higher than average accident rates among Hispanics in construction referenced in the literature, other variables will need to be developed in an attempt to isolate the differences between English and Spanish speaking employees. One possible explanation for no difference between the two groups could be due to the company chosen to participate in this study. The company chosen has a history of excellent safety scores. Dating back from 2004, the company has received safety awards from Associated General Contractors of America, the American Society of Concrete Contractors, and the Associated Builders and Contractors to name a few. A company publication lists their accident incident rate as well below the national average for their construction category. A company with an excellent safety record was chosen so significant differences could not be easily explained away by citing poor safety records or other extraneous factors involving poor safety. Regardless, at least in this national construction company, no significant differences in accident rates based on an individual's level of English proficiency exist. Being careful not to over-generalize the findings, it would appear if companies put as much importance on safety (Weidner et al., 1998) and safety culture as this one does, a reduction in the higher accident and injury rates currently seen among the Hispanic population might be the result.

Research question five tested if there was a difference between level of English proficiency and job hazards. The two groups were separated by their level of English proficiency. The middle group of participants on the level of English proficiency scale was removed in order to magnify the difference between the groups based on their level of English proficiency. The more Spanish group, N = 46, was made up of scores ranging from two–eight (the lowest possible score was four, so the two indicates the survey was only partially filled out). The more English group, N = 102, was comprised of scores ranging from 16–20. Based on these two groups, a slight difference in the mean score based on mean physical hazard existed. The more English group, with a mean of 3.226, perceived the worksite and conditions contained more physical hazards than their more Spanish-speaking counter parts.

The independent samples t-test for research question five, with both variables being normally distributed, was not significant. No significant difference between mean physical hazard scores based on level of English proficiency was present. The mean physical hazard variable looked at perception of worksite hazards possibly leading to injuries. Since no difference between the two groups was identified, it would appear both groups have similar views on physical hazards at the worksite.

Research questions seven through eleven identify whether significant correlations exist between the five independent variables used in this study (impulsivity, job tenure, safety knowledge, physical hazards, and level of English proficiency) and the dependent variable, overall injury score. Since overall injury score was not normally distributed, all correlations use the Spearman's Rho statistic instead of the more popular Pearson correlation statistic used for normally distributed variables. These research questions more closely mimic those presented by Frone (1998), as they do not differentiate on level of English proficiency, but seek only to identify if they are predictors of work injuries.

For research question seven, testing an association between impulsivity and work injuries, no significant correlation surfaced, as evidenced by Table 6, between mean impulse score and overall work injuries. A p value of .342 is well outside of the .05 necessary to indicate statistical significance.

Table 5. Spearman's Rho correlation for research question seven

		Sum of overall injury score
Mean impulse score	Correlation coefficient	−0.07
	Sig. (2-tailed)	0.342
	N	187

For research question eight, the test to identify an association between job tenure and work injuries, Table 6 indicates no significant correlation between job tenure and overall injury score. The significance score was .109, with 187 useable surveys out of the total of 191. Frone (1998) did not propose a hypothesis for the relationship between job tenure and injuries. Had a one-tailed test for significance been used, in lieu of the .05 for the two-tailed test, the resulting significance statistic would have come very close to the .10 cutoff. This would have mimicked the finding in the Frone study where it was found “job tenure was positively related to work injuries” (p. 573). This related to the literature discussed by Frone involving Iverson and Erwin's (1997) suggestion where more experienced workers would be given jobs necessitating greater skill, as well as being more dangerous to complete.

Table 6. Spearman's Rho correlation for research question eight

		Job tenure
Sum of overall injury score	Correlation coefficient	0.118
	Sig. (2-tailed)	0.109
	N	187

Research question nine, the test of association between safety knowledge and work injuries, did not produce a significant correlation between mean safety knowledge and overall injury score. The significance score was .352, with 186 useable surveys. Safety knowledge was not a variable directly linked by the literature to work injuries.

The data presented in Table 7, for research question ten, tested an association between physical hazards and work injuries. The table indicates a significant correlation between the two variables. With a significance value of .004, this correlation is significant at the .01 level. When the correlation coefficient of .216, indicating a small to medium effect size, is squared, the resulting value is .0467. With this value one can predict 4.67% of injury scores based on an individual's mean physical hazard score. The correlation also indicates as the mean physical hazard score goes up, meaning more participants believe their working conditions are not safe, so does the number of overall injuries.

Table 7. Spearman's Rho correlation for research question ten

		Mean physical hazard
Sum of overall injury score	Correlation coefficient	.216(∗∗)
	Sig. (2-tailed)	0.004
	N	178

Research question 11 (Table 8), identified a significant correlation between level of English proficiency and work injuries. The correlation coefficient was .172, a small effect size, with a significance value of .018. The resulting r squared value was .0296, indicating about 3% of work injuries can be predicted by an individual's level of English proficiency. In this correlation, as level of English proficiency increases, so does the overall injury number. This is directly opposite of the assumed relationship at the start of this study. This could have occurred by the underreporting of some injuries, or it could relate to the numbers present in each of the two groups. In any event, this is an interesting finding. Correlations could also be different if the study was conducted with a construction company lacking the high level of safety involvement the company used in this study contained.

Table 8. Spearman's Rho correlation for research question eleven

		Sum of level of English Prof.
Sum of overall injury score	Correlation coefficient	.172(∗)
	Sig. (2-tailed)	0.018
	N	188

Research questions 12 through 14 examine various regression models in relation to work injuries. Research question 12 tested whether a combination of impulsivity, job tenure, physical hazards, and safety knowledge predicts work injuries better than any one predictor variable in isolation. The dependent variable for this question is the overall injury score. From Table 9, three significant correlations within the regression model can be seen. The first is between mean physical hazard and overall injury score. This correlation was examined in research question ten. The Pearson correlation is .209, with a compound r square value of .0437 and a significance value of .003. As mean physical hazard scores increase, so does the overall injuries score. Mean physical hazard is also correlated with mean impulse score displaying a Pearson correlation of .179, a significance value of .008, and an r square statistic of .032. In this instance, as mean physical hazard scores increase, so does the mean impulse score. Both correlations indicate a small to medium effect size (Morgan, Leech, Gloeckner, & Barrett, 2004).

Table 9. Regression statistics for research question twelve

	Sum of overall injury score	Mean impulse score	Years working const.	Mean physical hazard	Mean safety know.
Pearson correlation
Sum of overall injury score	1.000
Mean impulse score	−0.086	1.000
Years working const.	0.062	0.033	1.000
Mean physical hazard	0.209	0.179	0.068	1.000
Mean safety knowledge	0.019	−0.296	−0.008	0.039	1.000
Sig. (1-tailed)
Sum of overall injury score	–
Mean impulse score	0.127	–
Years working const.	0.206	0.330	–
Mean physical hazard	0.003	0.008	0.183	–
Mean safety knowledge	0.399	0.000	0.455	0.303	–

The third correlation found in this regression model is the correlation between mean safety knowledge and mean impulse score. The negative sign in front of the Pearson correlation statistic of − .296 indicates a relationship where as impulse scores increase, the corresponding safety knowledge score decreases. The significance statistic is .000, indicating significance in the correlation of these two variables. The r square statistic is .0876 and the effect size for this correlation is considered medium. By knowing the mean impulse score of an individual, one can predict with 8.76% accuracy the associated mean safety knowledge score.

This model does not separate the participants based on their level of English proficiency. This first model had an overall R of .251, which would be considered a small to medium effect size. With an r squared statistic of .063, this model predicts 6.3% of overall work injuries based on the variables included.

The resulting ANOVA table Table 10, shows an F value of 2.887, and is statistically significant at the .05 level with a significance value of .024. The statistical significance indicates one or more of the variables are significant predictors of the dependent variable, overall injury score (Morgan, Leech, Gloeckner, & Barrett, 2004).

Table 10. ANOVA ^b table for research question twelve

	Sum of squares	df	Mean square	F	Sig.
Regression	452.379	4	113.095	2.887	.024 ^a
Residual	6737.361	172	39.171
Total	7189.740	176

Table 11 further identifies the effect of the different coefficients on the overall correlation coefficient. As can be seen by the significance value of .003 for mean physical hazard, the majority of the predictability can be explained by the effect of mean physical hazard on the regression model. Also contributing slightly to the overall correlation coefficient is mean impulse score with a significance statistic of .081, not significant at the .05 level, but adding to the overall correlation coefficient nonetheless.

Table 11. Regression coefficients (a) of research question twelve

	Unstandardized coefficients		Standardized coefficients
	B	Std. error	Beta	t	Sig.
(Constant)	2.121	3.146	–	0.674	0.501
Mean impulse score	−0.579	0.331	−0.138	−1.753	0.081
Years working const.	0.035	0.051	0.051	0.685	0.495
Mean physical hazard	2.013	0.657	0.232	3.067	0.003
Mean safety knowledge	−0.166	0.427	−0.030	−0.388	0.698

Research question 13, Table 12, looks at the same combination of impulsivity, job tenure, physical hazards, and safety knowledge to predict work injuries as did research question twelve, but further specifies the combination by only looking at individuals in the more Spanish group based on level of English proficiency. In this regression model, two correlations were statistically significant. One between mean physical hazard and mean impulse, and the other between mean physical hazard and years working construction.

Table 12. Regression statistics for research question thirteen

	Sum of overall injury score	Mean impulse score	Years working const.	Mean physical hazard	Mean safety know.
Pearson correlation
Sum of overall injury score	1.000
Mean impulse score	−0.122	1.000
Years working const.	0.035	0.234	1.000
Mean physical hazard	0.081	0.257	0.253	1.000
Mean safety knowledge	−0.011	−0.219	−0.016	0.233	1.000
Sig. (1-tailed)
Sum of overall injury score	–
Mean impulse score	0.213	–
Years working const.	0.409	0.061	–
Mean physical hazard	0.299	0.044	0.047	–
Mean safety knowledge	0.472	0.074	0.458	0.062	–

With a significance value of .044, the correlation between mean physical hazard and mean impulse is barely in the .05 range for significance. As is the case with the group regression model, the more impulsive an individual is, the lower their level of safety knowledge. Although the significance statistic is barely evident, the effect size of this correlation is small to medium with a Pearson correlation statistic of .257.

The other statistically significant correlation is between mean physical hazard and years working construction. This had a small to medium effect size also, as evidenced by the Pearson correlation of .253. As mean physical hazard scores went up, so did the corresponding levels of mean impulse and job tenure. The more job tenure a member from the more Spanish group has, the higher the perceived physical hazards are. This does make sense from the standpoint where the more you are exposed to dangerous things, the more you become aware of dangerous things.

The R for the overall model is .190, with an r squared of .036. Both identify this model as a poor predictor of work injuries. The resulting ANOVA shows no statistical significance either. To further reiterate this, none of the coefficients for this model are flagged as significant.

When the same model is run using the more English group in research question 14, the results are strikingly different. Here we are presented with three significant correlations (Table 13), two at the .01 level. The first is the correlation between mean physical hazard and overall injuries. The correlation coefficient is .295, indicating a small to medium effect, and the significance statistic is .001. The second significant correlation is between safety knowledge and mean impulse. The correlation coefficient is −.365, a medium to large effect size, with a significance statistic of .000. As impulsivity increases, the level of safety knowledge decreases, as evidenced by the negative sign. The more impulsive an individual is, the less likely they are to identify potential hazards.

Table 13. Correlations ^a for research question fourteen

	Sum of overall injury score	Mean impulse score	Years working const.	Mean physical hazard	Mean safety know.
Pearson correlation
Sum of overall injury score	1.000
Mean impulse score	−0.072	1.000
Years working const.	0.027	0.043	1.000
Mean physical hazard	0.295	0.091	0.031	1.000
Mean safety knowledge	−0.011	−0.365	−0.05	−0.179	1.000
Sig. (1-tailed)
Sum of overall injury score	–
Mean impulse score	0.237	–
Years working const.	0.393	0.334	–
Mean physical hazard	0.001	0.182	0.381	–
Mean safety knowledge	0.457	0.000	0.311	0.036	–

The third and final correlation is between mean safety knowledge and mean physical hazard. This correlation has significance at the .05 level with a significance statistic of .036. The correlation coefficient indicates a small effect size with a value of −.179. As the negative sign indicates, when mean safety knowledge increases, the corresponding level of physical hazards goes down. As one becomes more aware in relation to safety knowledge, the perception of physical hazards is not as great. This could lead to more injuries if the apparent safety knowledge makes individuals think things are safer than they actually are. On the flip side, as individuals gain in safety knowledge, they are able to complete dangerous tasks safely.

The overall model has a correlation coefficient of .312, a medium effect size, with an r squared value of .097. This would indicate almost 10 percent of work injuries among the more English group can be explained by this model. The p-value for this model is .041, just inside of the .05 cutoff. The mean physical hazard coefficient, at a value of .003, is the only significant coefficient in the model. This means the majority of the statistical significance for the overall model comes solely from the effect of mean physical hazard. This supports the almost intuitive notion, supported by Howe (2000); exposures to hazards cause injury.

Table 14 further breaks down the coefficients and identifies mean physical hazard, with a statistical significance of .003, as the primary contributor to the overall significance of the model. No other coefficients are significant contributors on their own. The significance of the mean physical hazard, based on the dependent variable of overall injuries, would indicate perception of hazards around the worksite contributes to higher overall injury scores. The more apparent hazards are, the more likely someone is to be injured.

Table 14. Regression coefficients ^a , ^b of research question fourteen

	Unstandardized coefficients		Standardized coefficients
	B	Std. error	Beta	t	Sig.
(Constant)	−2.067	6.895	–	−0.300	0.765
Mean impulse score	−0.847	0.905	−0.098	−0.936	0.352
Years working const.	0.016	0.067	0.023	0.233	0.816
Mean physical hazard	3.370	1.091	0.305	3.090	0.003
Mean safety knowledge	0.071	0.806	0.009	0.088	0.930

Auxiliary Findings

When Spearman correlations were run on most available variables in the study, several interesting correlations were identified. Only correlations with significance of .05 or stronger are included in this section, as can be seen in Table 15. The ∗ indicates significance at the .01 level.

Table 15. Other significant correlations of interest

Variables	Spearman correlation coefficient	Significance	N
Faces of pain & job tenure	0.182	0.013	187
English proficiency & level of school	0.575	0.000∗	182
English proficiency & faces of pain	0.179	0.014	187
Safety knowledge & level of school	0.151	0.044	180
Physical hazard & faces of pain	0.197	0.008∗	178
Company safety questions & job tenure	−0.166	0.024	185
Company safety questions & physical hazards	0.156	0.038	178
Overall injuries & faces of pain	0.328	0.000∗	184
Faces of pain & age	0.232	0.001∗	189
Age & physical hazards	0.151	0.045	178
Company safety questions & age	−0.211	0.004∗	183

There is a strong correlation between level of English proficiency and level of schooling. With a correlation statistic of .575 (a very large effect size), it appears the more English proficient employees are, the more likely they are to have completed a higher level of education. This was not the aim of the study, but is an interesting finding nonetheless.

Another correlation of note is one hoped to be present. The correlation of .328, a medium effect size with a significance statistic of .000, between the faces of pain scale and overall injuries shows the reporting of injuries to be fairly consistent. It shows participants were consistent with the number of injuries and how these injuries impacted their overall feeling of health. This also demonstrates convergent validity by the “relatively high correlations between [one] scale and other measures that the theory suggests would be related” (Gliner & Morgan, 2000).

Other correlations resulting from analyses show several other interesting relationships. Faces of pain and job tenure were correlated at a correlation coefficient of .182, a small to medium effect size, indicating the longer a person worked, the more likely they were to view themselves as having more pain. The correlation between English proficiency and faces of pain, correlation coefficient of .179, lends a bit more weight to the previous correlation between level of English proficiency and work injuries. In this instance, the higher the English proficiency, the more overall pain is felt. An interesting hypothesis here could be a difference between reporting of pain based on cultural differences. The level of pain could be quite similar, but one group could report less pain due to some cultural expectations or norms.

Safety knowledge is positively correlated with level of school, which agrees with the prior correlation between level of English proficiency and level of school. The more schooling an individual has, the higher the level of safety knowledge.

The prior correlation between physical hazards and overall work injuries is supported here by the correlation between physical hazards and faces of pain. Both seem to be indicators of physical hazards. As with the previous correlation between physical hazards and overall injuries, the higher the faces of pain score, the greater the perception of physical hazards.

The company safety question variable, developed using questions from the company's safety programs, was inversely related to job tenure. As fewer questions were answered no, indicating greater knowledge of the company safety questions, job tenure increased. Thus the longer a person is employed in this particular construction company, the greater their company safety knowledge. This would be an expected outcome, but is verified here. Company safety was also correlated to another variable, physical hazards. In this relationship the correlation statistic is .156 with a significance statistic of .038. Here, it appears the more you understand the company safety questions, the more likely it is you will perceive more hazards on the job site.

The correlation of the faces of pain scale to age is also an interesting finding. There was no correlation between overall injuries and age, but as age increases, so did pain as measured by the faces of pain scale. With a correlation statistic of .232, indicating a small to medium effect size, this is a fairly significant correlation.

Age was also correlated with physical hazards. As age increases, so does an individual's perception of physical hazards on the job site. This correlation was significant at the .05 level with a significance statistic of .045, and a correlation coefficient of .151 indicating a small effect size.

Mann Whitney U statistics, displayed in Table 16, were run including two more points on each side of the level of English proficiency continuum. The increase on the Spanish speaking side added more individuals closer to the bilingual designation. The English-speaking group had their scores lowered two levels, again adding more individuals closer to the bilingual midpoint.

Table 16. Mann-Whitney U test for main variables including more subjects

	Level of school	Years working const.	Faces of pain scale	Mean imp. score	Mean safety know.	Mean physical hazard	Mean of company safety questions	Sum of overall injury score
Mann-Whitney U	1076	3110.5	2777.5	3129	2868.5	3077	3108	2993
Asymp. Sig. (2-tailed)	0.000	0.115	0.023	0.17	0.039	0.673	0.339	0.088

The first significant difference, including more participants, is in relation to level of schooling. The correlation was previously identified involving the positive correlation between level of English proficiency and level of school completed. This difference in means further demonstrates the finding. Additionally, this difference is significant at the .01 level with a significance statistic of .000. A difference among the group means in the faces of pain scale also emerged. With a significance statistic of .023, this difference is significant at the .05 level. This finding further supports the earlier finding of the more English group indicating a higher overall injury score.

The third and final significant difference in means is found in mean safety knowledge. The more English group has the higher safety knowledge scores. With a significance statistic of .039, this difference is significant at the .05 level.

To further expound on the difference in level of schooling between the two groups, a regression was run with level of schooling and level of English proficiency as variables (see Table 17). The r value is .580 indicating a very large effect size. The resulting adjusted r squared value is .333; indicating 33.3% of an individual's level of schooling can be predicted by their level of English proficiency.

Table 17. Regression of level of schooling and level of English proficiency

R	R square	Adjusted R square	Std. error of the estimate
.580(a)	0.337	0.333	2.433

With the inclusion of two more points on each group relating to level of English proficiency, a regression model containing all relevant variables was run. The resulting model (Table 18), in an attempt to identify predictors of work injuries, identified an R of .155 (a small effect size) and an adjusted r squared value of 0.067 indicates 6.7% of work injuries can be explained by the variables in the model. The resulting ANOVA table identified the F statistic as 2.132 and the significance at .025, indicating significant predictors in the model.

Table 18. Regression of full model

R	R square	Adjusted R square	Std. error of the estimate
.355(a)	0.126	0.067	6.38016

In this model, only two of the coefficients were significant (Table 19). First was the faces of pain scale at a .007 level of significance, and the second was mean physical hazard at .015. Most of the predictability of the model is contained in these two coefficients.

Table 19. Coefficients of interest in regression of full model

	Unstandardized coefficients		Standardized coefficients
	B	Std. error	Beta	t	Sig.
Faces of pain scale	1.253	0.456	0.222	2.75	0.007
Mean physical hazard	1.857	0.755	0.198	2.459	0.015

Qualitative Findings from the Instrument

The final question of the research instrument was qualitative in nature. This was done to allow participants to voice any safety issues relating to their overall health, while working on the job, which may not have been addressed by the instrument. Twenty-seven of the 191 participants completed the qualitative question for a 14.14% response rate. Responses were grouped into categories. The 27 responses resulted in 15 categories and 29 pieces of information to place in a category. The categories and the number of times it was mentioned are included in Table 20. If a response contained multiple categories, it was counted for each category, hence the 29 pieces of information from 27 comments.

Table 20. Results of the qualitative question

Category	Number of responses
Air quality	10
Negligence of others as a safety concern	3
Education of others regarding safety	2
Noise levels	2
Stress	2
Teamwork leading to safety	2
Overhead traffic	1
Take care of your back by proper lifting	1
Pressure from a foreman resulted in accidents	1
Possible illegal material	1
Long strenuous hours	1
Language barrier	1
Hygiene (cleanliness of bathrooms)	1
Equipment causing soreness	1
Construction is just more dangerous	1

Of the 29 pieces of information, ten, or 34.45% related directly to air quality on the job. An example of one comment is a “constant battle of air particles and pollutants in work environments,” and “insufficient fresh air in the lower level.” Three individuals cited the negligence of others as their primary safety concern. Several other categories received two responses each. Those categories were stress, noise levels, teamwork leading to safer work environments, and education for certain groups relating to safety.

Modifications to the Instrument

In retrospect, several modifications to the instrument could improve the data collection process. Although developing the instrument to allow for scanning was effective in tabulation, a few areas were problematic. First, the questions relating to age, level of schooling, and hours worked, would have caused fewer problems had participants simply entered the appropriate number and not the corresponding bubbles. The researcher could then have entered those few pieces of data by hand. Second, clearer instructions might have prevented participants from filling the instrument out in crayon instead of pencil. This would have saved some time as it would not have been necessary to enter this data by hand.

In the groups observed by the researcher, monolingual individuals tended to be reluctant to ask questions. This was observed through behavior exhibited by the research study participants. In the researcher's opinion, if these groups were separated, with a Spanish-speaking individual providing more specific instructions to the group speaking only Spanish, more accurate data might have been obtained. One on one interpretation would have taken more time than the one half hour lunch break allotted by the company for completion of the survey tool.

Lastly, some of the qualitative data related specifically to dust and air pollution. Originally, the workers' primary occupation (i.e., painter, drywall, etc.) was left off of the instrument due to a human subjects concern. It would have been beneficial to know the primary occupation of the workers to identify if an issue affecting only painters or finishers was present, or if the problem was more widespread. With some care, there is no reason why this type of question could not be included in the instrument.

Conclusion

The overall findings of this research indicate no significant differences in work injuries between the following two groups: English proficient speakers and non-English proficient speakers. In this study, the more English proficient an individual was, the more likely they were to experience a greater number of injuries. Spanish speaking individuals having a higher accident rate than their more English speaking counterparts was contradicted in this study.

Areas for Further Research

These findings relate to this sample, for this national construction company, in one geographic area. These findings could be completely different when the same instrument is applied to other construction companies.

Since this study was primarily exploratory in nature, numerous areas for further research were found. Further study of significant differences in level of schooling, based on level of English proficiency, would be necessary to see if safety areas are affected. It would also be interesting to further expound on the inverse relationship between impulsivity and safety knowledge. Why does safety knowledge decrease as impulsivity increases?

One large area for additional research is to further examine the higher than average accident rates, cited by various authors, among Hispanic workers in the construction industry. Is there truly a difference between the groups, or is something else contributing to the difference? It is also important to recognize a difference in accident rates based on level of English proficiency and race could exist. More research will need to be done on this area in order to generalize the findings to greater populations. To the researcher's knowledge, this was the first study attempting to compare work injuries based on level of English proficiency as a predictor variable.

The correlation between level of English proficiency and work injuries could be further analyzed. Why, when the literature seemed to indicate Hispanics were being injured more often, did the data identify higher accident rates among the more English proficient group? Would the study have been better served to identify race in lieu of English proficiency? What would be the affect on literature which indicates the presence of language barriers, and their major contribution to accidents?

As exploratory studies have a tendency to do, far more questions have been generated than answered. It is important to recognize for this one company, when safety is taken seriously, level of English proficiency does not contribute to higher accident rates among more Spanish speaking individuals. To the researcher's knowledge, this was the first study attempting to compare work injuries based on level of English proficiency as a predictor variable. Future research should attempt to identify if similar findings are identified in companies where safety is not a top priority.

One last area for further research would be about the work injuries themselves. For this company, location and frequency of injuries has been documented. This could be an excellent starting point for research into preventive measures to reduce the most common work injuries in general; not just among certain populations.

Implications for Practice

Recommendations to be addressed from this study begin with the descriptive tables identifying location and type of injury. This could be of benefit to an organization interested in proactive solutions to ergonomic issues on a construction site. It could be a starting point for a wellness program, targeting several common injury areas, and demonstrating how to care for yourself so as to be less likely to sustain preventable injuries.

The literature cannot be unilaterally applied to all construction companies. As is the case with the company participating here, the higher accident rates are in the more English proficient group. The literature would lead one to believe anywhere non-English speaking individuals working in construction are present; their accident rates would be higher. It is entirely possible this approach could be misdirected. The problem could lie among the more English-speaking group, and the more English-speaking group could be the one to benefit from additional training.

Notes

Note. ∗∗Correlation is significant at the 0.01 level (2-tailed).

Note. ∗Correlation is significant at the 0.05 level (2-tailed).

Note. N = 177.

^a Predictors: (constant), mean safety knowledge, years working const., mean physical hazard, mean impulse score.

^b Dependent variable: Sum of overall injury score.

^a Dependent variable: Sum of overall injury score.

^a Selecting only cases for which sum of English proficiency = more Spanish.

^a Selecting only cases for which sum of English proficiency 2 groups = more English.

^a Dependent variable: Sum of overall injury score.

^b Selecting only cases for which sum of English proficiency = more English.

^a Predictors: (Constant), Mean of company safety questions, sum of level of English Prof., recode of hours worked last day, mean safety knowledge, faces of pain scale, mean physical hazard, years working const., mean impulse score, level of schooling, hours worked last day.