Impulsivity and the Sexes

Sex Differences in Impulsivity

The research literature on impulsivity has long acknowledged that sex differences in impulsivity exist, although few studies have addressed these differences systematically, choosing, instead, to examine the role of impulsivity in risk taking after controlling for the effects of sex. In general, it is usually recognized that males show higher levels of impulsivity, which is often associated with increased use of alcohol, nicotine, and caffeine (Waldeck & Miller, 1997), although this assumption does not always hold true. Costa, Terracciano, and McCrae (2001) found, using the NEO Personality Inventory–Revised (NEO-PI-R; a measure of Extraversion, Agreeableness, Conscientiousness, Neu-roticism, and Openness to Experience; Costa & McCrae, 1992), that men tend to be higher in excitement seeking (which is related to the sensation-seeking facet on the UPPS-P), whereas women scored higher on the impulsiveness facet (which is related to the negative urgency trait on the UPPS-P). No significant sex differences were found for self-discipline and deliberation facets (thought to be related to the lack of perseverance and lack of deliberation facets of the UPPS-P; Costa et al., 2001). Additional studies have failed to find significant differences between males and females on impulsivity (e.g., Patton, Stanford, & Barratt, 1995), including no meta-analytic effect size differences on the impulsiveness facet of the NEO-PI-R (Feingold, 1994). However, some evidence suggests that females make more effective use of delay of gratification than males and that this use is more advantageous in the long run (Silverman, 2003).

Reasons for sex differences in personality in general have mostly centered on biological, sociocultural, and biosocial models for sex differences (see Feingold, 1994, for a general discussion on sex differences in personality). Biological bases of sex differences suggest chromosomal differences between males and females that relate to differences in behavioral impulsive tendencies, sociocultural bases suggest that these differences are related more to social or cultural factors that form or condition these traits differentially in males and females, whereas biosocial bases suggest that biological and sociocultural influences interact to produce sex differences in personality (see Feingold, 1994, for a full discussion). Thus, much interest has existed in sex differences in impulsivity and the underlying mechanisms for these differences. It is important to determine whether or not males and females differ in their levels of impulsivity as it could shed light on mechanistic and risk factor differences between males and females, which could ultimately lead to more specific and effective treatment and prevention strategies.

Specific to impulsivity, some research has supported these underlying bases, whereas others have not. Some biological bases of impulsivity differences have been found. For instance, researchers have posited that these differences could be tied to higher binding capacity of 5HT(2) receptor in men than in women (Biver et al., 1996), which is related to increased impulsive action (Walderhaug et al., 2007), but this finding is not universally accepted. Other studies have failed to support differential roles of the interaction between biological and sociocultural effects for males and females in the development of impulsivity traits (e.g., Gunn & Smith, 2010). Notably, a recent study by Carver, Johnson, Joormann, Kim, and Nam (2011) found that the serotonin transporter polymorphism interacts with childhood adversity to predict increased levels of positive and negative urgency, regardless of sex.

A recent meta-analysis has once again brought to the forefront the issue of whether or not males and females exhibit different levels of impulsivity across multiple self-report and behavioral measures of impulsivity (Cross, Copping, & Campbell, 2011). Findings from 277 studies that included both an impulsivity measure and a measurement of participant sex indicate that men show higher levels of sensation seeking and behavioral risk taking, whereas women exhibit more punishment sensitivity; however, there was heterogeneity among the effect sizes found across studies (Cross et al., 2011). One reason for these inconsistent findings could be related to an unclear definition of what is meant by the term impulsivity. Research with sex differences using the UPPS-P traits could help clarify these discrepancies and add clarity to potential differential prediction of outcomes by sex. Additionally, one major limitation of this meta-analysis was the lack of establishment of measurement invariance across sex on the studied measures prior to the examination of group differences, thus questioning the meaningfulness of these findings. Therefore, the current study seeks to do these two things with one well-validated measure of impulsivity, the UPPS-P.

Measurement Invariance and Structural Invariance

Measurement invariance is determined to exist between two groups if the trait scores between the two groups are comparable and have the same meaning across the groups—in other words, whether or not the measure assesses the same trait in the same way across groups (Reise, Widaman, & Pugh, 1993). Until it is determined that a measure assesses the same trait across two separate groups, comparisons across the two groups on the measure have uncertain meaning (see Drasgow, 1984, 1987; Meade & Lautenschlager, 2004.). Methods to determine measurement invariance below are discussed below. However, if it is found that the measure is not invariant across groups, then comparisons between the two groups (e.g., on mean differences) are not meaningful, because group differences could be because of the different performance of the measure in the two groups (Reise et al., 1993; Windle, Iwawaki, & Lerner, 1998). Interestingly, despite considerable interest in possible sex differences in impulsivity traits, there has been little appreciation of the importance that measurement invariance plays in attempt to evaluate possible sex differences. Thus, although findings seem to suggest group differences on some aspects of impulsivity (see Cross et al., 2011), it is not clear whether these differences are because of actual trait-level differences or because of noninvariance of the psychometric properties measured.

To examine measurement invariance of a scale, multiple statistical tests and methods have been proposed. This procedure has been discussed in length elsewhere (see Byrne & Shavelson, 1987; Byrne & Watkins, 2003; Cheung & Rensvold, 2002; Drasgow, 1984, 1987; Hoyle & Smith, 1994; Meredith, 1993); here, I briefly outline the general procedures and hypotheses. One examines a series of increasingly stringent hypotheses of invariance, using either confirmatory factor analysis (CFA; e.g., in a structural equation modeling framework; Joreskog, 1971) or through item response theory (IRT; Lord, 1980). ¹ Measurement-level invariance is determined through the examination of the following aspects of measurement across the groups: (a) invariance of factor structure across groups (configural invariance), (b) invariance of factor loadings across groups (metric or “weak” factorial invariance), and (c) invariance of intercepts across groups (scalar or “strong” invariance; see Byrne, Shavelson, & Muthén, 1989; Cheung & Rensvold, 2002).

After the determination of measurement invariance, structural invariance is assessed. Structural invariance concerns whether or not the two groups differ in their distribution and/or mean. If measurement invariance holds, structural differences can be thought to be real and interpretable group differences. Structural invariance is determined by examining the following aspects of measurement across groups: (a) factor variance invariance and (b) factor mean invariance (see Chueng & Rensvold, 2002). Structural invariance cannot be adequately assessed until measurement invariance has been shown.

Little has been published about the measurement invariance of any impulsivity-related measures, including the UPPS-P. To date, some research has examined configural invariance and structural (mean) invariance of the UPPS-P, and this research will be reviewed next.

Configural Invariance: Evidence for the Existing Factor Structure of the UPPS-P Scale

The UPPS-P Impulsive Behavior Scale (Lynam et al., 2007) is a five-factor measure of dispositions toward rash action. Early results with the UPPS-R Impulsive Behavior Scale (which did not include positive urgency; Whiteside & Lynam, 2001) found that four separate, though related, first-order factors loaded on three different factors of the five-factor model: lack of deliberation and lack of perseverance loaded negatively onto a factor that included NEO-PI-R Conscientious facets, sensation seeking loaded with NEO-PI-R facets measuring Extraversion, and (negative) urgency loaded onto NEO-PI-R facets measuring Neuroticism (Whiteside & Lynam, 2001). Additional studies have supported four separate first-order factors in both undergraduate and adolescent samples (D’Acremont & Van der Linden, 2005; Van der Linden et al., 2006; Magid & Colder, 2007; Smith et al., 2007; Whiteside et al., 2005; Whiteside & Lynam, 2003), sometimes with a higher order domain related to deficits in conscientiousness (as in Smith et al., 2007). When the UPPS-R was expanded to include the Positive Urgency Measure (PUM; from Cyders et al., 2007) and was renamed the UPPS-P (Lynam et al., 2007), factor analyses supported the presence of five separate, though related, first-order factors in undergraduates (Cyders & Smith, 2007) and young adolescents (Gunn & Smith, 2010), with three higher order factors: an overall deficit in conscientiousness factor (lack of deliberation and lack of perseverance as subfactors), a sensation-seeking factor, and an urgency factor (with positive urgency and negative urgency as subfactors).

In general, there is consensus that the UPPS-P measures five separate, though related, dispositions toward rash action that are best understood as represented by three second-order factors: lack of conscientiousness, sensation seeking, and emotion-based dispositions (Cyders & Smith, 2007). However, research has yet to determine whether or not this factor structure is invariant between males and females, despite a large interest in how impulsivity traits might differ between males and females, as discussed above. Therefore, this analysis will test the invariance of this five-factor measure with three second-order factors across males and females for configural invariance, as a first step of the examination of measurement invariance across these groups.

Structural (Mean) Invariance: Sex Differences on the UPPS-P Traits

Research that has examined sex differences in impulsivity using the UPPS-R or the UPPS-P has offered some clarification on the heterogeneity of sex differences findings that have been found in the literature (e.g., Cross et al., 2011). This research makes it is clear that sex differences in impulsivity depend on the specific disposition toward rash action that is measured. In general, males self-report higher levels of sensation seeking (correlation between sex and sensation seeking: r = 0.33, p < .001, in D’Acremont & Van der Linden, 2005; r = 0.25, p < .001, and r = 0.24, p < .001, for a questionnaire and interview version, respectively, in Cyders & Smith, 2007), whereas there are no sex differences in self-reported lack of deliberation (r = 0.04, p = .33, for D’Acremont & Van der Linden, 2005; r = −0.04, p = .46, and r = 0.40, p = .58, for a questionnaire and interview version, respectively, in Cyders & Smith, 2007) or lack of perseverance (r = 0.00, p = .99, for D’Acremont & Van der Linden, 2005; r = −0.07, p = .19, and r = 0.00, p = .98, for a questionnaire and interview version, respectively, in Cyders & Smith, 2007). The picture of negative urgency is less clear. Although some research has supported higher rates of negative urgency among females (r = −0.14, p < .001, in D’Acremont & Van der Linden, 2005), other research has found mixed support for this relationship (r = −0.03, p = .95, and r = −0.15, p < .05, for a questionnaire and interview version, respectively, in Cyders & Smith, 2007). To date, only one study has addressed whether or not sex differences in positive urgency exist; this study reported no sex differences in positive urgency (r = 0.06, p = .27, and r = −0.12, p < .88, for a questionnaire and interview version, respectively, in Cyders & Smith, 2007).

Differential Prediction of Outcomes by Sex

Miller, Flory, Lynam, and Leukefeld (2003) were interested in how the relationship between impulsivity traits and risk taking might be modified by sex. They demonstrated differential correlations between the UPPS-R facets and risk outcomes between males and females for the following relationships: urgency and the variety of conduct problems (males, r = 0.31, p < .001; females, r = 0.25, p < .001) and borderline symptoms (males, r = 0.08, p = n.s.; females, r = 0.34, p < .001); sensation seeking with borderline symptoms (males, r = −0.11, p = n.s.; females, r = 0.08, p = n.s.); lack of perseverance with crime and delinquency (males, r = 0.26, p < .001; females, r = 0.46 p < .001), alcohol use (males, r = −0.06, p = n.s.; females, r = 0.26, p < .001), number of sexual partners (males, r = −0.07, p = n.s.; females, r = 0.15, p < .05), and age of sexual debut (males, r = 0.17, p < .05; females, r = −0.12, p = n.s.); lack of deliberation and crime and delinquency (males, r = 0.32, p < .001; females, r = 0.52, p < .001).

In general, there is high interest in how males and females might differ in impulsivity and the implications of such possible differences. However, this research is premature and has begun to examine construct-level invariance before first establishing measurement invariance on assessment procedures. To be clearer, one explanation for sex differences in impulsivity could be that males and females have different levels of impulsivity traits; this hypothesis has mixed support in the literature. An alternative explanation is that the items on the UPPS-P are answered differently by males and females or that the items translate into different levels of the trait at various response levels (both questions of measurement invariance across the groups). Therefore, one cannot answer questions concerning group differences or differential prediction of a trait without first examining measurement invariance across the groups, both at the scale and individual item levels.

The Current Study

Given the high interest in impulsivity group differences between males and females and the coexisting paucity of research examining measurement invariance in impulsivity across males and females, the current study seeks to examine both measurement invariance and construct invariance of one well-validated measure of impulsivity, the UPPS-P, across males and females. This study used a CFA in structural equation modeling in MPlus (Asparouhov & Muthén, 2009), examining for psychometric measurement invariance by testing the following: the examination of (a) invariance of factor structure across groups (configural invariance), (b) invariance of factor loadings across groups (metric or “weak” factorial invariance), and (c) invariance of thresholds across groups (scalar or “strong” invariance; see Byrne et al., 1989; Cheung & Rensvold, 2002). After the determination of measurement invariance across groups as described above, structural invariance was assessed, by examining the following: (a) factor variance invariance and (b) factor mean invariance (see Chueng & Rensvold, 2002). Finally, relationships between the UPPS-P traits and outcome criteria traditionally of interest in impulsivity research for invariance in these predictor–outcome relationships were examined for differential prediction.

Method

Results

Measurement Invariance Across Sex

To examine the incremental measurement invariance hypotheses, as presented earlier, a CFA approach in structural equation modeling was conducted with MPlus 5.2.1 (Asparouhov & Muthén, 2009). Weighted least squares mean- and variance-adjusted estimation was used, as the UPPS-P items are ordered categorical data. The following fit indices for the analyses are reported: the comparative fix index (CFI; Bentler, 1990), the Tucker–Lewis index (TLI; Tucker & Lewis, 1973), and the root mean square error of approximation (RMSEA; Browne & Cudeck, 1993). Guidelines for what constitutes a good fit vary, although a CFI greater than either .90 or .95, a TLI equal to or greater than .96, and an RMSEA equal to or less than .06 are thought to represent good fit for categorical data (Schreiber, Stage, King, Nora, & Barlow, 2006). Women served as the reference group in all invariance models.

Configural invariance

A hierarchical three-factor model of the UPPS-P was examined for configural invariance, separately in males and females. Individual items were used as indicators of latent factor variables, and all latent variables were allowed to intercorrelate. Reverse-coded items were allowed to have correlated error residuals. Factor loadings and item thresholds were freely estimated in both groups. Item residual variances were fixed to 1 and factor means were set to 0. Factor variance was fixed to 1 for identification. This model served as the baseline configural model for subsequent analyses. Overall, this hierarchical model fit the data well, although not ideally: χ²(df = 241) = 2531.55, p < .001; CFI = .86; TLI = .96; RMSEA = .11.

When tested separately, the model fit similarly for males and females: χ²(df = 187) = 2113.20, p < .001; CFI = .83; TLI = .96; RMSEA = .10 for females; and χ²(df = 40) = 7238.14, p < .001; CFI = .85; TLI = .95; RMSEA = .12 for males. Although the CFI had slightly better fit for males than females, the RMSEA showed the opposite pattern, and since the TLI was unchanged between the groups, it was determined that, relatively speaking, the models fit equally well in each group. Additionally, estimates between the two groups were did not appear to be markedly different. Thus, there was sufficient evidence to support configural invariance across males and females, despite the overall slightly lower than ideal fit with the CFI and the RMSEA fit indices.

Of note is that the use of parcels with this model significantly improved the fit of the model, especially pertaining to the CFI and the RMSEA, χ²(df = 160) = 661.49, p < .001, CFI = .95, TLI = .94, RMSEA = .06 for females; and χ²(df = 160) = 473.99, p < .001, CFI = .94, TLI = .93, RMSEA = .07 for males, thus replicating the support of this model in previous studies, many of which also used parceling, ² and to exhibit the effect of the number of individual item indicators of fit indices (although much of this support has been within maximum likelihood estimation; see Cook, Kallen, & Antmann, 2009; Kenny & McCoach, 2003; Marsh, Hau, Balla, & Grayson, 1998); however, individual items were retained for invariance testing in the current study because of the controversial nature of parceling and for the clear clinical utility of examining the individual items for measurement invariance analyses. Additionally, although the fit of this overall model was not optimal and was likely deflated, due, in part, to the use of individual item indicators, the goal of invariance testing is to test change in model fit with increasing constraints, and thus modification indices that suggested cross-loadings of items were examined but were not retained in the final model because (a) individual sample modification indices are likely not to be robust (MacCallum, Roznowski, & Necowitz, 1992) and (b) when they were applied to the model in an exploratory analysis, CFI model fit improved, but only slightly, whereas RMSEA model fit decreased. Finally, given the large sample size of the current study, more weight was placed on the fit of the TLI, which appears to be relatively independent of sample size (although this is better established in the maximum likelihood estimation framework; see Marsh et al., 1998). Therefore, it was determined that the application of these modification indices would introduce more uncertainty to the model with very little benefit to the overall fit indices, especially within the context of a well-fitting TLI value.

Thus, a series of model constraints were then applied to the baseline model, which used individual item indicators, in a successive fashion to examine potential decreases in fit resulting from lack of measurement or structural invariance. Fit indices for this model and for the models tested in the remaining measurement invariance hypotheses are presented in Table 1. The ΔCFI test was used to evaluate fit changes with increasing constraints, using a criterion of ΔCFI < .01 to indicate no significant reduction of fit, as suggested by Cheung and Rensvold (2002).

OPEN IN VIEWER

Table 1.

Measurement Invariance Summary Fit Statistics

Model	χ2	χ2 degrees of freedom	CFI	TLI	RMSEA estimate	ΔCFI	χ2 difference test (significance value)
Configural invariance model (baseline model)	2531.55	241	.863	.96	.11	NA	NA
Metric invariance model	2586.01	272	.861	.96	.10	−.002	NA
Scalar invariance model	2694.39	279	.855	.96	.10	−.006	NA
Factor invariance model	2208.52	256	.883	.97	.10	NA	3.18 (.36)
Full-factor mean invariance model	2201.58	246	.883	.96	.10	NA	69.36 (<.001)
Partial-factor mean invariance model (sensation seeking released)	2139.75	247	.886	.97	.10	NA	21.06 (<.001)
Partial-factor mean invariance model (positive urgency released)	2165.89	252	.885	.97	.10	NA	11.69 (.003)
Partial-factor mean invariance model (lack of perseverance released)	2178.46	254	.885	.97	.10	NA	4.09 (.09)

Note. CFI = comparative fix index; TLI = Tucker–Lewis Index; RMSEA = root mean square error of approximation; NA = not applicable. All χ² values were significant at p < .001.

Metric invariance

Equality of the unstandardized item factor loadings across groups was then examined in a metric invariance model in which the factor loadings were estimated freely in women but were in men set to equal the freely estimated parameters from the women. Individual items were used as indicators. The factor variance was fixed to 1 in women but was freely estimated in men, and the factor means were fixed to 0 in both groups. Residual variances were freely estimated in each group. The metric invariance model did not result in significantly decreased fit as compared with the configural invariance model (see Table 1), ΔCFI < .01. The fact that metric invariance held indicates that the items related to the latent factor equivalently across groups or, more simply, that the same latent factor was being measured in each group.

Scalar invariance

Equality of the unstandardized item thresholds across groups was then examined in a scalar invariance model. The factor variances and means were fixed at 1 and 0, respectively, for identification in the women, but the factor variance and mean were then estimated in the men. All factor loadings and item intercepts were freely estimated in females but constrained to be equal to females in the male group; all residual variances were still freely estimated in both groups. The scalar invariance did result in a significantly decreased fit of the model over the metric invariance model (see Table 1), ΔCFI < .01. This finding suggests a lack of response bias between the groups and allows comparisons of factor means across groups to be made.

Structural Invariance Across Sex

Thus, since measurement invariance was found to hold between males and females, a series of tests was conducted to examine the structural invariance of the UPPS-P between males and females. To compare these latent parameters, a chi-square difference test (Δχ²) was examined.

Factor variance invariance

The equality of factor variances was now examined in a model in which factor variances were set to be equal across groups. Factor means were set to 0 in women but were allowed to be freely estimated in the male group. This constraint did not significantly lower the fit of the model (see Table 1), Δχ² = 3.18, df = 3, p = .36. This finding supports factor variance invariance across sex.

Factor mean invariance

Next, the equality of factor means across groups was tested by constraining the factor means to be equal across groups. This model did have a significantly lower fit than the previous model (see Table 1), Δχ² = 69.36, df = 3, p < .001. Therefore, factor means were released singularly, and change in fit was reevaluated. Modification indices suggested that releasing factor mean for sensation seeking still resulted in a significantly worse fit (Δχ² = 21.06, df = 2, p < .001). Modification indices next suggested releasing the mean of positive urgency; this also resulted in a significant chi-square difference test (Δχ² = 11.67, df = 2, p = .003). Modification indices next suggested releasing lack of perseverance; this resulted in a nonsignificant chi-square difference test (Δχ² = 4.09, df = 2, p = .09). Independent sample t tests somewhat supported this finding, showing that in a subset of the participants, males report significantly higher levels of positive urgency and sensation seeking than females (see Table 2). Although nonsignificant, the difference between males and females on lack of perseverance was trending toward higher levels in males (see Table 2).

OPEN IN VIEWER

Table 2.

Mean Levels and t Tests for Impulsivity Traits by Males and Females

Trait	Mean	SD	t Statistic	df	Significance
Positive urgency			−5.80	1,272	<.001
Female	1.68	0.54
Male	1.87	0.59
Lack of deliberation			−0.94	1,272	.35
Female	1.98	0.44
Male	2.00	0.44
Negative urgency			0.08	1,272	.94
Female	2.32	0.58
Male	2.32	0.57
Sensation seeking			−9.32	1,272	<.001
Female	2.85	0.56
Male	3.15	0.51
Lack of perseverance			−1.47	1,272	.14
Female	1.81	0.44
Male	1.85	0.42

Note. Females n = 827, males n = 447. Female is coded as 1 and male is coded as 2, so negative t-test statistics indicate higher levels of the trait for males.

Differential Relationships of Impulsivity Traits to External Outcomes by Sex

Given the findings of full measurement invariance and higher rates of positive urgency and sensation seeking in males (and potentially lack of perseverance), differential prediction of the impulsivity traits by sex was next examined, using a subset of the overall sample to analyze how the impulsivity traits might be differentially related to risk outcomes, by sex.

First, using a subsample of n = 418 participants, we examined the relationships among the impulsivity traits and (a) drinking frequency, (b) drinking quantity, (c) drinking problems, (d) positive emotion–based risk taking, (e) negative emotion–based risk taking, and (f) general risk taking. This subsample consisted of first-year college students, with a mean age of 18.52 (SD = 8.08). In all, 25% (n = 105) of the sample was male. A total of 88% percent of the sample was European American (n = 368), 8% African American (n = 35), and 4% was of another race (n = 15).

Correlations among the impulsivity traits and sex in this sample are presented in Table 3. Overall, correlations among the impulsivity traits replicated previous findings and ranged from r = 0.62 (between negative urgency and positive urgency) to r = −0.11 (between lack of perseverance and sensation seeking). Sex was significantly correlated with sensation seeking (r = 0.18, p < .001), lack of perseverance (r = 0.14, p < .01), and positive urgency (r = 0.15, p < .01).

OPEN IN VIEWER

Table 3.

Bivariate Correlations of Study Variables (n = 1,274)

	SS	NUR	PUR	LD	LPS	Sex
SS		0.13**	0.20**	0.26**	−0.07*	0.25**
NUR			0.57**	0.36**	0.37**	−0.00
PUR				0.31**	0.33**	0.16**
LD					0.49**	0.03
LPS						.04
Sex

Note. SS = sensation seeking; NUR = negative urgency; PUR = positive urgency; LD = lack of deliberation; LPS = lack of perseverance. Traits measured using the Positive Urgency Measure and the UPPS-R. Values here include imputed values from total sample of n = 1,274. Values were unchanged using nonimputed data.

*

p < .05. **p < .01.

Correlations among the impulsivity traits and the risk outcomes by sex, for the subset of the total sample, are presented in Table 4. A series of z tests were performed comparing the correlations among the traits and risk outcomes between males and females (see Table 5). Overall, results indicate that, for the variables measured, the relationships between the self-reported traits and the risk outcomes did not differ by sex, with one exception: There was a significant difference between lack of deliberation and general risk taking (z = −2.27, p = .02). Additionally, these correlations appear to support the convergent and discriminant validity of the impulsivity traits, replicating previously reported correlational patterns. For example, as previously reported, sensation seeking, negative urgency, and positive urgency all related to drinking quantity (see Cyders et al., 2009), whereas lack of perseverance and lack of deliberation were unrelated. All impulsivity traits were significantly related to drinking frequency and drinking problems, which does replicate other findings, although other research focusing on the incremental utility of these traits suggests that specificity of prediction occurs not at the bivariate level but at the incremental level, which controls for the effects of the other traits (see Cyders et al., 2009, for an example of this). As found in other studies, positive mood and negative mood–based rash action was bivariately related to all the traits, although other research has found specificity in incremental validity (e.g., Cyders & Smith, 2010).

OPEN IN VIEWER

Table 4.

Correlations Among the Impulsivity Traits and Sex in Sample Subset (n = 418)

	Sex	LD	SS	LPS	NUR	PUR
Sex		−0.01	0.18**	0.14**	0.03	0.15**
LD			0.25**	0.39**	0.34**	0.31**
SS				−0.01	0.19**	0.25**
LPS					0.36**	0.34**
NUR						0.62**
PUR

Note. SS = sensation seeking; NUR = negative urgency; PUR = positive urgency; LD = lack of deliberation; LPS = lack of perseverance. Traits measured using the Positive Urgency Measure and the UPPS-R. Female was coded as 1 and male was coded as 2, so positive correlations indicate higher levels of the trait for males.

*

p < .05 **p < .01.

OPEN IN VIEWER

Table 5.

Correlations and Differences Among Impulsivity Traits and Risk Outcomes by Sex in Sample Subset (n = 418)

	SS	LPS	NUR	PUR	LD
DQ
Males	0.40**	0.17	0.38**	0.31**	0.16
Females	0.32**	0.10	0.25**	0.23**	0.29**
z statistics (p)	0.79 (.43)	0.61 (.54)	1.24 (.22)	0.74 (.46)	−1.18 (.24)
DF
Males	0.36**	0.24*	0.38**	0.31**	0.31**
Females	0.30**	0.03	0.20**	0.24**	0.29**
z statistics (p)	0.58 (.56)	1.85 (.06)	1.70 (.09)	0.65 (.52)	0.19 (.85)
DP
Males	0.22**	0.21*	0.35**	0.30**	0.29**
Females	0.14*	0.14*	0.35**	0.35**	0.25**
z statistics (p)	0.71 (.48)	0.62 (.53)	0.00 (1.00)	−0.48 (.63)	0.37 (.71)
POS
Males	0.38**	0.20*	0.45**	0.32**	0.19
Females	0.26**	0.10	0.33**	0.38**	0.33**
z statistics (p)	1.15 (.25)	0.88 (.38)	1.22 (.22)	−0.59 (.56)	−1.29 (.20)
NEG
Males	0.31**	0.20*	0.49**	0.38**	0.33**
Females	0.15**	0.18**	0.39**	0.31**	0.26**
z statistics (p)	1.46 (.14)	0.18 (.86)	1.07 (.28)	0.68 (.50)	0.25 (.51)
GEN
Males	0.45**	0.12	0.41**	0.43**	0.17
Females	0.45**	0.07	0.27**	0.33**	0.41**
z statistics (p)	0.00 (1.00)	0.43 (.67)	1.37 (.17)	1.01 (.31)	−2.27* (.02)

Note. SS = sensation seeking; NUR = negative urgency; PUR = positive urgency, LD = lack of deliberation; LPS = lack of perseverance; traits measured using the Positive Urgency Measure and the UPPS-R. DQ = drinking quantity, DF= drinking frequency and DP = drinking problems, assessed by the Drinking Styles Questionnaire; POS = positive emotion–based risk taking and NEG = negative emotion–based risk taking, as measured by the Mood-Based Questionnaire; GEN = general risk taking, as assessed by the Risky Behavior Questionnaire. All items were coded so that higher levels correspond with more risk-taking behavior. z Statistics compared correlation values between males and females (two-tailed). Female was coded as 1 and male was coded as 2, so negative z statistic indicates higher levels among females.

*

p < .05. **p < .01.

Discussion

The current study sought to meet three aims: (a) to examine psychometric measurement invariance of the UPPS-P for males and females, (b) to examine the structural invariance across males and females, and (c) to examine how the impulsivity traits might differentially relate to risk outcomes, as predicted by sex. These questions are important as they speak to the appropriateness and the validity of using the UPPS-P to measure trait levels of the dispositions toward rash action and to predict risk for males and females. In fact, although research has previously suggested trait-level differences between males and females on impulsivity, determination of whether or not there are trait differences could not be fully answered until measurement invariance between the two groups was established. This is the first study to conduct a comprehensive evaluation of the use of the UPPS-P to measure self-report impulsivity for measurement invariance, construct invariance, and differential criterion-related validity.

The findings generally indicate that the UPPS-P is a valid and reliable scale for the measurement of impulsivity traits for both males and females. In other words, the UPPS-P can be administered and scored as recommended by its authors for both males and females; no modifications are necessary. However, the results of this study should be replicated with other samples to determine the validity of the UPPS-P in, for instance, clinical samples. Additional outcomes of interest should be examined for differential prediction as well. The current study also suggest that although validly assessed across males and females, it appears that males report higher levels of both positive urgency and sensation seeking than females, although these mean-level differences do not translate into differential prediction of risk outcomes, in general. The findings also suggest that males likely have higher levels of lack of perseverance than females, although data from the current study do conflict to some extent and this should be replicated in other research.

More generally, this study supports the usefulness and importance of examining the validity of using and comparing the results of questionnaires across groups. Much interest exists in the field of psychology concerning group differences, including not only males and females but also group differences based on race, socioeconomic status, and sexual orientation (e.g., Reynolds, 1983). Researchers and practitioners alike must appreciate the need to establish measurement invariance across groups in order to have confidence in the meaning of purportedly substantive differences in group means (see Cleary, Humphreys, Kendrick, & Wesman, 1975). Testing for measurement invariance is, unfortunately, rarely done (see Vandenberg & Lance, 2000). This has been recently seen in the case of impulsivity research: A recent meta-analysis that found group differences in impulsivity (Cross et al., 2011), in short, should be understood and interpreted within the context of measurement invariance. Until measurement invariance can be found to hold between males and females on the wide array of measures assessed in that study, a conclusion concerning the existence of differences between males and females is premature. Therefore, conclusions posited by the authors (Cross et al., 2011) could reflect trait-level differences, as the authors suggest; however, these findings are also consistent with the existence of measurement noninvariance across males and females, and thus, this should be examined empirically.

Additionally, examination of measurement invariance is especially important in multicultural research or in industrial/organizational psychology, where construct validity limitations of a measure can result in group differences that are psychometric in nature (e.g., related two noninvariant factor loadings) but are interpreted as substantive group mean differences and potentially biased prediction, treatment assignment, hiring, promotion, or diagnosis. This psychometric test bias can result in biasing because of metric noninvariance or scalar noninvariance. If metric invariance does hold but scalar invariance does not, the groups cannot be compared, since the constructs themselves account for the mean differences across the groups. The use of different cutoff scores could be used to better predict for or assign individuals from different groups (see Cleary et al., 1975; Cole, 1981). It is important to note that this solution is not without controversy, and is often illegal; different cutoff scores for groups should only be used with caution and only when the measure demonstrates clear scalar noninvariance (see Cleary et al., 1975; Cole, 1981). One more widely accepted solution (although still not free of controversy) in industrial/organizational psychology is known as score banding, in which scores within a certain range are treated as equivalent (see Campion et al., 2001), with the goal of more representative hiring or promotion decisions.

However, if scalar invariance holds, but metric invariance does not, the construct factor itself is different. In this case, different predictive relationships would exist between the predictor and the criterion based on group membership (see Cleary et al., 1975). Again, this type of bias could be overcome with the use of different regression equations (with different weights) for prediction, as long as both relationships are positive, although, similar to above, this could be quite controversial. This type of bias is known as psychometric test bias (see Cleary et al., 1975) and is thought of as extremely problematic for any measure or item holding this characteristic. The measure could be potentially used to study within group differences, but any analysis of between group differences using this measure would be considered differentially valid or biased. Use of the same regression weight across groups would not be appropriate, and use of the measure at all across groups could be considered unsuitable.

In conclusion, the current study supports the measurement invariance of the UPPS-P across males and females and suggests that comparisons in impulsivity levels (as assessed by the UPPS-P) across males and females can be validity interpreted as representing underlying mean differences in the traits. The current study clarifies research concerning sex differences in impulsivity, suggesting that males and females differ in sensation seeking and positive urgency facets of impulsivity, as well as possibly in lack of perseverance, thus clarifying some of the research inconsistencies in this area. Importantly, these mean differences in traits between males and females do not translate in differential prediction of a variety of risk outcomes assessed.