<<

Markers of Adulthood Subscale Development 1

Markers of Adulthood Subscale Development:

Comparative Review and Assessment of Inductive and Deductive

Nathan E. Fosse1

Harvard University

Jon E. Grahe

Alan Reifman

March 10, 2015

1 Direct all correspondence to Nathan E. Fosse at [email protected]. Please cite as follows: Fosse, N., Grahe, J. E., & Reifman, A. (2015, March 31). Markers of Adulthood Subscale Development: Comparative Review and Assessment of Inductive and Deductive Psychometrics. Retrieved from https://osf.io/p8nwq/

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 2

Executive Summary

In this report we outline two psychometric approaches to develop subscales from the 40 Markers

of Adulthood items in the Emerging Adulthood at Multiple Institutions (EAMMI) survey. First

we describe the inductive approach, drawing from empirical analyses of 20 of the items. Second

we describe the deductive approach toward scale construction. In addition to describing the

processes of scale construction, we also describe the attributes and psychometric properties of

the scales derived. Finally, we assess the inter-construct validity of inductive vis-a-vis deductive approaches, finding high correlations between the two sets of measures. In addition, we draw from use of the 40 MoA items in the extant literature to provide a comparative analysis of the

EAMMI items. We find that, despite divergent 'styles' of psychometric reasoning, both approaches are consistent with the strengths and he challenges of constructing scales in the literature. We conclude with a review of the lessons learned. The appendix includes details of the

PCA analyses run by the first author, describing the findings in greater detail.

Keywords: Markers of Adulthood, emerging adulthood, deductive reasoning, psychometrics, principal components analysis.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 3

Markers of Adulthood Subscales in the EAMMI Survey

In this methodological supplement we describe two psychometric approaches we had

undertaken in constructs to construct scales from derived from 40 Likert-type Markers of

Adulthood (MoA) items in the survey. We undertook two approaches toward the data, comparing findings, assessing measured by internal reliability (i.e. by the consistency or

homogeneity of items comprising subscales) and theoretical coherence (the degree to which subscales make theoretical sense given current understanding of emerging adulthood). We adopted a sample-based, bottom-down, deductive approach as well as a literature-based, top-up inductive approach toward scale measurement. We describe the attributes and the internal validity of each subscale derived, and we compare both approaches empirically and conceptually. We find that both approaches to capture the most important aspects in the MoA items and see future benefits in collaborative psychometrics.

While each item assesses the responses of participants to 20 adulthood markers, and may be treated as single indicators, we recognize the potential advantages of advancing emerging adulthood research through dedicated investigation of theoretically-relevant and empirically- valid scales.

We undertook different approaches toward generating MoA scales in part out of necessity

(that is, the need to produce parsimonious and communicable findings) but also because there is not widespread consensus in constructing scales from the MoA items. The present analyses and description of approaches describe features of the EAMMI data, providing important supplemental analyses of the sample structure. In addition, the present analyses provide some templates and guidelines for future research using similar items.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 4

The analysis is present in sections that follow. First we provide a synopsis of the EAMMI

items. In the next two sections, we describe the data and the two approaches we took in

constructing MoA subscales, based in the psychometric literature: inductive approaches, which

begin with sample-based data analyses to inform theoretically-meaningful scales, and; deductive

approaches, which begin with prior literature and conceptual distinctions and only secondarily to

the sample characteristics. The former approach can be considered bottom-up, as it focuses on the data and analytics, while the latter is top-down drawing from the literature to generate informative priors to generate scales. In practice, however, both approaches take into consideration both prior knowledge and the findings at hand.2 We describe the process of scale

construction, along with the internal reliability and descriptive statistics of the measures.

Comparing both approaches based on comparisons of internal subscale reliability and of

conceptual or theoretical coherence, we conclude similarity in the constructs derived.

Fourth, we compare each of the subscales we generated with each other empirically, and

with respect to prior analyses selected by N. Fosse for their conceptual advances. Based on both

empirical and theoretical comparisons, we conclude that the constructs derived match in broad

conceptual ways, but we also see promise in advancing he emerging adulthood literature in

collaborations that address the following points: (1) outlining in detail the approaches toward

scale construction benefit from collaboration, particularly across social scientists with different

“styles of thought” (Hacking, 1994); (2) theory and methodological orientations reveal how scale

construction can produce different forms of knowledge, particularly without consensus-making among collaborators through sharing syntax code, and analytic decisions; (3) theory-building is

2 A third approach common in psychometrics, the criterion approach, assumes a priori the important criteria for condition or participant classification, and develops scales based on their prediction of this external standard. In contrast, both approaches we describe are internal, as they are based on the properties of the data itself.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 5 intrinsic to the practice of psychometrics. In the emerging adulthood literature, there is much scientific knowledge to be gained from further investigation of emerging adulthood(s), particularly as this life stage expands and transforms across global society. As a psychological and cultural life stage (Arnett, 2011), it will continue to be crucial to understand the object of inquiry as we also seek to understand its variation across social milieu.

Synopsis of EAMMI Markers of Adulthood Items

The EAMMI survey contains 40 questionnaire items assessing how important each of 20

Markers of Adulthood (MoA) are important in achieving adulthood and whether the respondents achieved these adulthood identifiers. All items and assessing each of the 20 MoAs were derived from Arnett’s analyses (1997, 2001).

This analysis concerns 40 items covering two conceptual domains. First, 20 measure participants’ perceived importance (1 = very important, 4 = not important) of obtaining each

MoA. Second, 20 items measured achievement of each MoA (dichotomous, yes or no; or ordinal, from 1 = very true, 3 = not true). Also included in the EAMMI survey is single item measuring whether a respondent felt he or she had reached adulthood (yes, no, somewhat). This factor variable was recoded so that higher scores represented fuller attainment of adulthood. As a global measure of attainment unrelated to any particular MoA, this item was not included in subscale constructions.

Following the conceptual distinction adopted in psychometrics (Burisch, 1984; Smith,

Fischer, & Fister, 2003), we characterize the subscales we generated from the items the result of two different scale construction processes: inductive and deductive approaches. Inductive approaches rely on information from the data itself, i.e., the correlations or covariances between

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 6

items, to reduce data empirically-meaningful clusters. Theoretical coherence is assessed during

the process of analysis, based on both the interpretability of findings but also on assessments in

light of prior literature. In contrast, deductive approaches or orientations generate constructs

based on exiting literature and theory first, and test internal validity based on those a priori theories.

We proceeded as follows. First, J. Grahe computed a principal components analysis

(PCA) on the 20 importance items, presenting these subscales to potential authors prior to the invitation to participate in the Special Issue. To incorporate both importance and attainment items, N. Fosse adopted a ‘top-down’ deductive approach, presenting to other authors memoranda and soliciting feedback from corresponding authors. N. Fosse revised subscales based on further review of prior literature, for conceptual clarity, and for communicability. Other authors were invited to create their own subscales to share or to revise these constructs. We describe these two approaches in the following sections, respectively.

Inductive Approach (Sample-based, Empirical)

The primary advantage of inductive approaches, or sample-based, bottom-up psychometrics, is that it permits the introduction of new constructs that have not been explored or considered by previous theory. In addition, bottom-up approaches provide empirical support for assessing the structure of the data. The main limitation, however, is that inductive statistics can rely on assumptions that are not warranted in some data, producing uninformative results. In addition, subscales that fit data may have little theoretical import However; we mitigated these possible limitations by comparing analyses with prior literature and by replicating analyses across statistical platforms.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 7

J. Grahe applied PCA using SPSS to the 20 importance items in the EAMMI dataset, and

found four factors representing 53 % of the total variance. N. Fosse replicated these analyses in

Stata/SE 13 finding additional evidence to retain the first four factors. Specifically, N. Fosse ran

a principal components analysis based on the correlation matrix of the 20 importance items,

interpreting the factors from their contribution to variance. Factor loadings may be transformed

in different ways, although results should be substantively similar. N. Fosse produced factor

loadings of the first four retain factors, also using varimax rotation. The descriptive statistics of the 20 importance items used by N. Fosse in Stata/SE are located in Table 1 of the appendix.

As additional check, N. Fosse examined whether all 20 importance items should be included in PCA estimation. PCA requires at least moderately correlated variables to produces informative results. N. Fosse ran two pre-estimation analyses. First, he ran a Kaiser-Meyer-Olkin

(KMO) test of sampling to determine whether the variables have enough variance in common to warrant PCA. KMO ranges from 0 to 1.0, where smaller values indicate variables have too little in common to warrant PCA. The average KMO across all 20 items was .86, warranting PCA on all items, and considered ‘meritorious’ (Kaiser, 1974). Second, N. Fosse conducted the squared multiple correlation (SMC or ) of each item against all other variables in the analysis. Based on 2 the values, some items may𝑅𝑅 warrant exclusion. However, lower R2 should be considered a 2 conservative𝑅𝑅 estimate. This is in part since PCA is based on the assumption that each item is

parametric and continuous (i.e., a normally-distributed ratio or an interval variable). In short, the pre-estimation KMO and SMC tests warrant inclusion of all 20 items, though they suggest some caution given the non-parametric, non-continuous structure of the items. The KMO and 2 statistics for each of the 20 items are located in Table 2 of the appendix. 𝑅𝑅

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 8

Supporting the analytic decision made by J. Grahe to keep the first four components of

the PCA, N. Fosse’s replicated analyses also revealed that 53 % of the variance in the items is

explained by four components. Detailed output is located in Table 3 of the appendix, revealing that the first component explains 27% of the variance in the 20 items.

The scree plot derived from N. Fosse’s analysis support the four-component selection, where criteria depend on subjective identification of the bend or the ‘elbow’ in the plot. Based on this criterion, the scree plot suggests that three factors could be retained.

An additional criterion for component selection, provided in Figure 1 supports retention of four components. In particular the horizontal line in the scree plot, is the demarcation of the

Kaiser criterion, which is to drop any eigenvalue >1.0, or the amount of variance explained by an average item (Jolliffe, 2014).

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 9

Figure 1.

Scree Plot of Eigenvalues of 20 Importance Items.

Note. Scree Plot based on principal components analysis of 20 importance items in the EAMMI

survey (n = 1,309) using Stata/SE 13.0.

While there is also no consensus regarding interpretation of principal components in

PCA, it is common to rotate orthogonal components. While rotating destroys properties of the initial components, relaxes the assumption that each component is uncorrelated with any other

(i.e., that they are orthogonal). The most common technique is varimax rotation, which maximizes the sum of the within-column variances.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 10

Table 1 shows the factor loadings with each item, derived from analyses by J. Grahe in

SPSS. N. Fosse also replicated these findings in Stata/SE, which are included in the appendix in

Table 3. Loadings from analyses by N. Fosse using Stata/SE are reported as eigenvalues, but separation of loadings revealed identical component structures to those found using SPSS.3

J. Grahe assigned the following labels to the factors, matching conceptual distinctions in

Arnett’s work. The labels of each subscale in decreasing order from Factor 1 to Factor 4 were, respectively: Family/Career (8 items); Drugs and Alcohol (4 items); Emotional Independence (5 items); and Financial Independence (3 items). Because missing data in single items caused missing data in the factor scores, mean scores using each set of items was also computed for the full sample.

In addition to assessing the theoretical coherence of each subscale, J. Grahe examined the validity of mean scores instead of weighted PCA factor scores, J. Grahe ran Pearson R correlations between the four components and mean scores derived from items. Correlations between factor and mean scores were high: (Family/Career, r [1307] = .95; Drugs/Alcohol, r

[1306] = .95; Emotional Independence, r [1307] = .93; and Financial Independence, r [1307] =

.83; all ps < .000). Given the high correlations between the factor components and the mean scores, it was decided to leverage the full sample of participants by using mean scores.

3 Stata/SE factor loadings are shown as eigenvectors, which are loadings normalized so that each column sums to 1, which is conventional in mathematics, and the default in Stata/SE 13.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 11

Table 1.

Item Loadings of Principle Components Analysis of EAMMI MoA Importance Items.

Variable Component 1 2 3 4 67. Married 0.84 0.10 0.01 0.02 68 Have at least one 0.84 0.06 0.01 -0.07 73 Long-term relationship 0.74 0.20 0.12 0.10 69 Long term Career 0.72 0.09 -0.03 0.23 83 Capable Support Parents 0.61 0.10 0.23 -0.05 76 Capable Caring Children 0.56 0.05 0.28 0.33 66 Finished Education 0.55 0.13 0.01 0.32 75 Capable Supporting Family 0.55 0.04 0.22 0.47 71 Avoid Illegal Drugs 0.20 0.80 0.08 0.15 70 Avoid Becoming Drunk 0.25 0.74 0.08 0.01 79 Avoid Drunk Driving -0.07 0.64 0.30 0.20 72 Use Contraception 0.09 0.63 0.08 0.13 80 Interact with Parents as Equals 0.20 0.04 0.66 0.03 82 Become Less self-oriented 0.18 0.25 0.64 -0.07 81 Control Your Emotions 0.27 0.18 0.59 -0.05 74 Make Independent Decisions -0.15 0.03 0.56 0.42 77 Accept Responsibility for Actions -0.23 0.11 0.55 0.38 64 Financially Independent 0.12 0.09 0.01 0.68 65 No Longer Live with Parents 0.06 0.12 0.01 0.61 78 Be Employed Full Time 0.34 0.21 0.13 0.55 Note. Factor loadings derived from a varimax PCA with retention of 4 principal components using SPSS. Loadings > .3 presented in bold. Mean scores were derived from items with the highest loading in each column.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 12

Detailed Description of PCA Subscales

Table 2 presents descriptive statistics on the mean scores, their items, and the Cronbach’s

for each unweighted mean score. The means, standard deviations, and reliabilities of this

construct,𝛼𝛼 and others are available on the Table of Overall Means on the Materials Component

on the EAMMI OSF project page (https://osf.io/b7ngc/).

The results in Table 2 below provide more granular information on the shape of the

subscale distribution, providing the median, skewness, and kurtosis for each mean score. For

skewness, positive values indicate skewed-right distributions, and negative values indicate

skewed-left distributions. Approximately symmetric distributions are between -.5 and .5; values

above -1 and +1 are considered highly skewed . Absolute values between .5 and .1 reveal

moderately skewed distributions. Kurtosis indicates the relative shape of the values of each

variable’s ‘peak’ compared its tails. Higher values indicate that variability in the data is located

in the extrema of the distribution. Values > 3 indicate excess kurtosis (Balanda & Macgillivray,

1988).

Importance mean scores indicate left skew for Factors 2, 3, and 4, and they indicates

excess kurtosis, revealing distributional variation at the extreme tails of these measures; the sample variability, in other words, is located in the tails of the distribution for these scores.

Reliabilities are acceptable by conventional standards.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 13

Table 2.

Description of Mean Scores of Importance Items, Derived from Inductive Approach.

Component Descriptive Statistics

n M SD Mdn. Skew Kurt. Items

Factor 1: Family/Career 1338 2.43 0.76 2.50 -0.14 2.08 8 𝛼𝛼.87

Factor 2: Drugs and Alcohol 1338 1.99 0.77 2.00 0.58 2.62 4 .72

Factor 3: Emotional Independence 1337 1.59 0.47 1.60 0.98 4.29 5 .64

Factor 4: Financial Independence 1336 1.70 0.60 1.67 0.71 2.99 3 .55

Note. All 20 importance items share the same Likert scaling in the EAMMI data set, ranging from 1 (‘very important’) to 4 (‘not at all important’).

This section we have described the use of an inductive approach toward he construction

of importance ratings of 20 markers of adulthood. Principal components analysis conducted in

SPSS and replicated in Stata/SE indicate the justification for selection of four subscales, a

description of items used to derive mean scores, and, finally, a table of descriptive statistics

providing insight into the asymmetric distribution of these measures, as well as their reliabilities

and attributes. From the items used to construct these measures, EAMMI contributors have

constructed conceptually-similar items using the 20 attainment items. The reliabilities and

statistics of these mean scores are available on the Table of Overall Means on the Materials

Component on the EAMMI OSF project page (https://osf.io/b7ngc/).

Deductive Approach (literature-based, theoretical)

In addition to the principle components analytic approach, N. Fosse approached the problem independently and established scales based on prior knowledge and theory, deriving analyses from distinct conceptual domains from the emerging adulthood literature. This approach

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 14 follows the approach taken by Arnett (1997), who developed six conceptual domains, relying on a pilot studies, earlier published research, and expansive, inter-disciplinary review of literature on the transition to adulthood.

After cleaning and coding the 40 measures of Markers of Adulthood using Stata/SE, N.

Fosse shared the constructs initially derived a memorandum with other authors in June of 2014 soliciting feedback and remarks; in addition, he consulted his coauthor regarding decision points at each step. With additional review of the literature and for conceptual clarity, N. Fosse proposed a revised subscale construction, drafted in November of 2014 in another memorandum, which he shared with collators for feedback. In addition, assessment of constructs relied on internal reliability measures.

Motivation for Deductive Approach

Deductive approaches are beneficial when analyses are theory-driven and when there is considerable a priori knowledge regarding the structure of subscales; in the language of Bayesian statistics, deductive approaches are beneficial when the analyst has stronger and more informative priors regarding the expected structure of data (Gelman, 2004). Properly-derived and informed subscales have been argued to permit more communicable and parsimonious constructs (DeVellis, 2011).

There are statistical reasons to place emphasis on deductive or top-down psychometric processes. In particular, several articles have described and provided examples of the widespread

“misuse” of the coefficient in (Green & Yang, 2008; Sijtsma, 2009). Latent variable models are proposed𝛼𝛼 by Green & Yang (2008), yet these models require advanced statistical training, which may lead to further misapplication and interpretation. For example,

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 15

widespread misapplication of latent measurement multilevel models are demonstrated to provide

little information for model selection (Nezlek, 2011).

An additional problem arises when attempting to reduce the dimensionality of data that are coded and asked in heterogeneous ways. In such instances, principal components analysis

(PCA) is uninformative and potentially misleading, as results are sensitive to outliers and are less

or non-informative when there are non-parametric, non-continuous indicator items (Jolliffe,

2014). PCA is unsuitable as well for the attainment measures. Since PCA attempts to explain all variation in the data indicator items, results are misleading when the model includes differently-

constructed indicator variables. In particular, four of the 20 attainment items in the EAMMI

survey are binary (1 = ‘yes’, 2 = “no”), while the remaining 16 attainment items were scaled at

three factor levels: 1 = ‘very true’, 2 = ‘somewhat true’, 3 = ‘not true.’

The difficulty of reducing the existing items, combined with considerable research on emerging adulthood, provide incentives to develop constructs based on theoretic grounds. This process of developing scale items from prior research first, in conjunction with psychometric analysis of scale items, is described below, in five steps. This process was iterative and included several revisions by N. Fosse as he learned more of the data.

Clean and Import the Data to Learn of the Items.N. Fosse imported the original

EAMMI survey in January 2014 from Excel into Stata/SE, and then in June and July of 2014. He used the questionnaire to create syntax that: (1) labeled variables using the codebook verbatim,

(2) assign item values labels, (3) run descriptive statistics to check for coding error and anomalies in the data set. N. Fosse exported the dataset from Stata/SE to SAS to other authors and editors.

Create subscales and identify need for theoretical matching of itemsIn January 2014

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 16

N. Fosse relied on the subscales from the PCA analyses and generated constructs, but found in

June of 2014 that some of the attainment items did not match the constructs the literature, that

they were coded differently than other measures, and that reliability could be improved by

excluded or rearranging some indicators. He shared the memorandum of these constructs with

the editors and with authors. In an updated memorandum, he provided revised constructs based

on paper revisions. At each of these steps N. Fosse examined the characteristics of the items,

constructed total as well as mean scores, and calculated reliabilities.

Identify prior use of items in the literature.The first step employed was to examine the

items in the literature, starting with some of the most highly-cited work by Arnett’s (1997, 2001).

In addition, other item subscales were derived from research based on exploratory factor

analyses (Badger, Nelson, & Barry, 2006; Nelson et al., 2007).

Select subscales based theoretical development. N. Fosse matched most of the EAMMI

MoA items into three domains based on theoretical development in the literature.

First, ‘role transitions’ was derived from Arnett’s (1997) use of items in the EAMMI data.

Second, the subscale ‘norm compliance’ derived from Badger et al. (2006). Third, ‘relational ’ is a construct derived from analyses by Nelson et al. (2007).

All three constructs provide theory development through conceptual clarity. First, the

‘role transitions’ construct matches measures of traditional demographic markers of the transition to adulthood (Schoon, 2015). ‘Norm compliance,’ conceptualizes risky behavior as a

developmental stage embedded within cultural context. Finally, ‘relational maturity’ offers a

construct that, like ‘norm compliance’ and role transitions, is theorized within the human

developmental framework. Conceptually, the three constructs are meaningfully linked, higher

scores on each of the three measures corresponds to the attainment of markers of adulthood:

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 17

increased attainment of role transitions, norm compliance, and relational maturity. This approach

provided constructs from 14 of the 20 items for each marker of adulthood.

Develop a meaningful construct from the remaining items.From the six remaining

items, there was less conceptual consistency across analyses. Arnett (1997) combined some items under the category of ‘responsibility’, while Badger et al. (2006) and Nelson et al. (2007) used some of these item to describe a construct called ‘family capacities.’

Combining across analyses, N. Fosse categorized the remaining 6 markers of adulthood items that measured aspects of ‘responsibility.’ This construct is less clear, as evinced by the diverse use of items in the literature, yet, it provides a construct that encompasses the different uses of these items. As an analytic construct, ‘independence’ describes items that assess both financial responsibilities as well as capacities to care for others. Higher scores on this measure, as do the others, indicate greater accomplishment of adulthood.

Recode items so they match the theoretical constructsThe MoA items include 40 items in total, one set of 20 that assess importance of attaining the criteria, and another set of 20 that assess whether or not participants attained those markers. N. Fosse undertook the following steps to construct items for the subscales.

Group Matched Importance Items to Identify Recodes. N. Fosse examined the

questionnaire to match each item analogous attainment achievement item, organizing each item

along each construct. Matched items were included in a spreadsheet for modification for

different specifications of subscales as well as for documentation. Following this protocol, N.

Fosse recoded variables if they were logically part of the subscale, but differed in its coding.

Reverse Code Items so Higher Values Indicate Greater Importance and

Attainment.Because all 40 importance items are derived from the same Likert-type scale, N.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 18

Fosse reverse-coded all in Stata so that higher values indicated greater endorsement of each

marker. The remaining 14 items for the attainment measures were reverse-coded to take on the

following values: 1 = ‘Not true’, 2 = ‘Somewhat true’, 3 = ‘Very true.’

Create Role Transitions Scale. Because 3 of the 6 items in the Role Transitions

Attainment subscale were binary, N. Fosse recoded those items so that a value of ‘1’ indicates

attainment of that marker, and ‘0’ indicate its contrapositive. N. Fosse reverse-coded item 86 and

recoded the three-level item number 90 “Have settled into a long-term career?” So that ‘1’ is equal to ‘Very, Somewhat true,’ and 0 indicates ‘Not true.’ Reliabilities were also assessed with the Kuder-Richardson coefficient of reliability, which is preferred for reliabilities of scales with binary items.

Based on revisions and iterative analyses, N. Fosse shared with the authors and editors his subscale derivation in November 2014. He ran the output, generated a subset of the MoA mean scores and items, transferred these data from Stata/SE to SPSS, providing output to be validated by K. Oleson and for analyses to be shared among other authors.

Detailed Descriptive Statistics and Reliabilities

Descriptive statistics of the subscales are available on the Table of Overall Means in the

Materials Component on the EAMMI OSF project page (https://osf.io/b7ngc/). Detailed features

of the subscales, including the median, skewness, kurtosis, and range, describe the shape of each

measure not presented in the overall table linked above.

Table 3 reveals that, similar to the PCA results, most cases were included in the sample,

with only 15 to 16 participants dropped for non-response to each full set of subscale items. In

addition, they fall within conventional standards in psychometrics, with values of about .60 or

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 19

above. Skew is approximately the same, with (with values reversed in direction due to reverse

coding of variables). Absolute values of skewness are somewhat smaller, with two variables

falling into the moderate skew categorical (i.e., with absolute values between .5 and 1). Kurtosis is somewhat smaller, with the exception of Independence, which is > 3.0. Nevertheless, kurtosis is moderately high enough to warrant caution with regard to interpretation of mean values, as variation is also located in the tails of the distribution.

Table 3.

Description of Mean Scores of Importance Items, Deductive Approach.

Mean Scores Descriptive Statistics

n M SD Mdn. Skew Kurtosis Items α

Role transitions 1337 2.64 0.7 2.5 0.19 2.22 6 .77

Norm compliance 1338 3.01 0.77 3 -0.58 2.62 4 .72

Relational maturity 1336 2.98 0.6 3 -0.22 2.64 4 .58

Independence 1337 3.29 0.49 3.33 -0.6 3.16 6 .65

Note. All items range from 1 to 3. Sample sizes differ because mean scores are not calculated if

all items are missing for participants.

Table 4 shows the sample statistics, along with the maximum and minimum of each

subscale. Role transitions are comprised of only binary variables, while the other three subscales

are comprised of three-level factor variables. Skewness is high for role transitions indicating that most of the EAMMI participants do not complete role transitions, and that most variability is located in the right tail of the distribution. The other three measures have moderately high

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 20 kurtosis, but moderate or low skew. Again, these findings imply that transformations of the measures would improve fit from regression models.

Reliability is greater than .60, indicating conventional standards, with the exception of the ‘relational maturity’ construct, which is below psychometric standards. Future work would benefit from more precise alternate measures. In the present analysis, the inter-item covariances reveal no potential increase in reliability by excluding items. Details on the attainment of relational maturity items are located in Table 5.

Table 4.

Description of Mean Scores of Attainment, Deductive Approach

Mean Scores Descriptive Statistics

M SD Mdn. Min. Max. Skew Kurtosis Items α

Role transitions .22 .23 .17 0 1 1.41 4.64 6 .70

Norm compliance 2.54 0.43 2.75 1 3 -0.81 2.91 4 .61

Relational maturity 2.18 0.42 2.25 1 3 -0.06 2.48 6 .38

Independence 1.89 0.39 1.83 1.17 3 0.8 3.24 6 .70

Note. n = 1,338 for all items. Reliability for the Role Transitions subscale was also calculated using the Kuder-Richardson coefficient (KR20 = .72).

The constructs reveal expected internal consistencies with the exception of the ‘relational maturity’ attainment measure. The remaining seven subscales are close, but not meeting standard criteria of = .70 (Bland & Altman, 1997) or of .60 (DeVellis, 2011, p. 109). As pointed out by

Badger et al.𝛼𝛼 (2006), the reconfiguration of conceptual domains with empirical findings warrant investigation, as prior analyses have found reliabilities as low as .42 and .53 (Arnett, 2003; Barry

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 21

& Nelson, 2005). The aforementioned analyses provide comparable constructs across both

objective as well as subjective items of emerging adulthood.

Given the low reliability of the Relational Maturity construct, Table 5 shows the

attributes of that subscale. In particular, it is important to examine the source of low reliability.

In particular, mean scores assume each item contributes equally to the construct. each item contributes identically to the overall measure. In contrast, PCA extracts components as linear combinations of weighted factors as a function of each item’s contribution to the variance in the

set of items included in the model.

There are several important insights from Table 5. First, column 1 shows the sample size for each item, revealing full coverall of the EAMMI participants. Specifically, Column 2 is the

correlation of each item with the overall alpha of .39. Column 3 examines the correlation of each

item with each item and the scale constructed without that item. Column 4 shows the average

inter-item correlation; higher values indicate higher correlations between items excluding the

item in the row. Column 5 provides the teste scale alpha if that item were excluded.

Findings show that the construct reliability cannot be reliability improved. In particular,

Column 1 shows that most participants are included in the item construction. Second, Columns 2 and 3 should be similar across all items, assuming that the analyst may generate each score as an unweighted sum of the items themselves. Finally, Columns 4 and 5 suggest that the items themselves have low reliability because they are correlated little with each other, and that removing any one item will not improve the scale reliability. On average, the inter-item correlation is .07. While low reliability is not necessarily important, the findings suggest treating each of these items as distinct outcomes with different theoretical implications.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 22

Table 5.

Item properties of Attainment of Relational Maturity Subscale.

Item 1 2 3 4 5

n test rest inter α

94 Committed to a long-term love relationship 1328 .67 .18 .07 .39

101 Have relationship with parents as equals 1328 .59 .26 .06 .27

102 Learned to always control emotions 1334 .56 .23 .07 .31

103 Become less self-oriented 1333 .57 .20 .07 .33

Test Scale .07 .39

Note. Subscale is derived from mean scores of items described in the first column. Column 1 indicates the number of participants responding to each item, column 2 is the item-test correlation of each item, column 3 is the item-rest correlation of each item, column 4 is the inter- item covariance for each item, and column 5 is the reliability measure if that item were excluded in the subscale.

The deductive approach undertaken by N. Fosse and outlined in five steps with construct generated here reveal a number of advantages. It is largely derived on a priori theory and on recent research findings, both of which permit testing of reliabilities as a priori tests. Also, these constructs adhere closely to the initial conceptual typology and matches with traditional conceptualizations of ‘role transitions’ in demography, which typically examines the ‘big five’ transitions of leaving home, completing one’s education, getting married, having a child, and settling into a full-time career (for a review, see Shanahan, 2000).

Inductive vis-à-vis Deductive Scale Constructions

In practice, both inductive and deductive approaches are used in scale construction, in particular since most clustering and factor analytic techniques used in psychology require the analyst to make analytical and conceptual distinctions based on interpretations of the data. That

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 23 is, there are is no universal ‘stopping rule’ for determining the number of latent components in a matrix of items (Jolliffe, 2014). In turn, deductive approaches may be revised based on reassessments of hypothesized constructs against the data.

The aim of this section is to examine the empirical and conceptual similarity of the derived constructs. First, to examine the degree of validity between constructs, we present pairwise correlations of the mean scores derived from each approach, located in Table 5. Higher correlations indicate that constructs are assessing similar measures.

Second, to compare the conceptualization of the MoA scores, we present results PCA estimated by J. Grahe on the 20 importance items, and we have the conceptual sources of items used in other selected studies on markers of adulthood in previous literature compiled by N.

Fosse and used to derive his initial conceptual model.

In Table 5 we present analyses of the mean scores from the deductive importance items

(reverse-coded to provide interpretive clarity) and the mean scores from the inductive importance items. Subscales are numbered from (1) to (4), based on the factor numbers from the PCA. In addition, the inductive subscales are indexed by the letter ‘a’ while the deductive scores are indexed by the latter ‘b’. Deductive mean scores were matched based on conceptual overlap.

Correlation coefficients to assess inter-construct validity are indicated by boxes. Correlations greater than .50, described the literature as a ‘moderate’ to ‘large’ or ‘substantial’ association, are presented in bold (Kotrlik, Williams, & Jabor, 2011).

The results in Table 6 support the hypothesis that both PCA and deductive mean scores measure similar constructs. The correlations between deductive measures and analogous factor loadings were very high: (1b) and Family/Career, r [1337] = .91; (2b) with Drugs/Alcohol, r

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 24

[1338] = 1.00; (3b) with Emotional Independence, r [1336] = .80; and, (4b) with Financial

Independence, r [1336] = .56. all ps < .000). The first three principal components reveal the

highest correlations among subscales, with the Drugs/Alcohol and the Norm Compliance

subscales sharing identical items.

The correlations outside the primary statistics of interest are smaller between items are

smaller. However, all correlations are statistically significant, with ps < .000. A related question

is the conceptual distinction of the inductive and deductive measures. The first column of Table

5 showed a high correlation of Family/Career with two deductively derived subscales: Relational

Maturity (r[1336] = .67) and with Independence (r[1337] = .72). This reveals the conceptual

breadth of this construct. In a similar manner, the deductively-derived Role Transitions measure

is correlated with the PCA-derived Financial Independence Measure (r[1336] = .59). In

summary, the findings in Table 6 support the hypothesis that each set of subscales exhibit high inter-construct validity.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 25

Table 6.

Pairwise Correlations of Deductive and Inductive Measures.

a. Inductive b. Deductive

Index Variable (1a) (2a) (3a) (4a) (1b) (2b) (3b) (4b)

(1a) Family/Career 1.00

(2a) Drugs/Alcohol .36 1.00

(3a) Emotional Ind. .30 .37 1.00

(4a) Financial Ind. .40 .33 .31 1.00

(1b) Role Transitions .91 .35 .27 .59 1.00

(2b) Norm Compliance .36 1.00 .37 .33 .35 1.00

(3b) Relational Maturity .64 .43 .81 .33 .55 .43 1.00

(4b) Independence .72 .39 .56 .56 .57 .39 .57 1.00

Note. All ps < .0000. Variables based on pairwise Pearson R correlations. Variables are mean scores derived from the important items in the EAMMI survey, with variables (1b) to (4b) reverse-scored for clarity. Sample size ranges from n = 1336 to 1338.

Cross-Study Consistency of MoA Important Subscales

The following section underscores the empirical findings in Table 6. The findings reveal cross-analysis consistency of the use of EAMMI items as indicators of Role Transitions. The

PCA shows that all except one of these items loaded on the first principal component. The first component is somewhat broader than the constructs derived in the literature, with some items included in the PCA construct that are not included in the other reference.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 26

Norm compliance is another construct with high inter-study consistency. Across the PCA

analysis, the construction crated by N. Fosse, and the items used in the remaining three articles,

norm compliance includes identical items.

The third construct derived by N. Fosse, Relational Maturity, reveals some inter-study agreement, yet this measure less clearly conceptualized than the Role Transitions and the Norm

Compliance constructs. As shown in Table 5, of the four items in the Relational Maturity construct derived by N. Fosse, 3 items correspond to the Emotional Independence construct

(Factor 3) from the PCA analysis, 3 correspond to the Emotional conceptual domain described by Arnett (1997), while no items match the concept of relational maturity in Badger et a. (2006).

In contrast, three of the items correspond to the construct derived from Nelson et al. (2006).

Aside from the analyses by Badger et al. (2006), the Relational Maturity construct reveals some agreement that this construct or some similar measure encompasses socioemotional development. However, conceptual ambiguity, combined with the low reliability of this measure revealed ( = .39) reveal that the items comprising this construct require greater conceptualization.𝛼𝛼 It was found from the reliability analyses in Table 4 that these items exhibit low correlations, suggesting they measure divergent aspects of the same phenomenon, or something else altogether.

The last subscale, Independence, shows divergent conceptualizations in the literature, incorporating at times the following constructs: responsibilities, cognitive development, role transitions, family capacities, role maturity, and relational maturity. The PCA reveal these items to be associated with the Financial Independence measure primarily. Nevertheless, as Figure 1. revealed, PCA analyses show that this construct explains less of the variance in the 20

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 27 importance items than the first three components. As an analytic construct, despite its high reliability ( = .70), this construct requires further theoretical conceptualization.

This𝛼𝛼 final section supported the correlation analyses in Table 6, which revealed high inter-method reliability. In particular, they showed high consistency across studies in the items used to construct measures of normative measures of the transition to adulthood and absolute agreement in the items to assess norm compliance. However, more subjective criteria assessing socioemotional development, or Relational Maturity, as well as measures of Independence, require further theoretical and empirical refinement. To the extent that the MoA items capture relevant constructs using the MoA items, the differences in the inductive and deductive approaches reflect the state of the emerging adulthood literature.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 28

Table 6.

Comparison of Conceptual Domains for Importance Items

Item PCA Fosse (2015) Arnett (1997) Badger et. al (2006) Nelson et al. (2006) 65. No longer living at parents 4 Role Trans. Role Trans Role Trans Role Trans 66. Finished with education 1, 4a Role Trans. Role Trans Role Trans Role Trans 67. Married 1 Role Trans. Role Trans Role Trans Role Trans 68. Have at least one child 1 Role Trans. Role Trans Role Trans Role Trans 69. long-term career 1 Role Trans. Role Trans Role Trans Role Trans 78. Be employed full-time 1, 4 Role Trans. Role Trans Role Trans Role Trans 70. Avoid becoming drunk 2 Norm Comp. Behavioral Norm Comp Norm Comp 71. Avoid illegal drugs 2 Norm Comp. Behavioral Norm Comp Norm Comp 72. Use contraception 2 Norm Comp. Behavioral Norm Comp Norm Comp 79. Avoid drunk driving 2 Norm Comp. Behavioral Norm Comp Norm Comp 73. Committed long-term relationship 1 Relational Mat. Emotion Independence N/A 80. parents as an equal 3 Relational Mat. Emotion Independence Relational Mat. 81. control of your emotions 3 Relational Mat. Emotion Independence Relational Mat. 82. Become less self-oriented 3 Relational Mat. N/A Independence Relational Mat. 64. Financially independent 4 Independence Role Trans. Independence Role Mat. 74. Make independent decisions 3, 4 Independence Cognitive Independence N/A 75. capable of supporting a family 1, 4 Independence Responsibilities Family Capacities Family Capacities 76. capable of caring for children 1, 4 Independence Responsibilities Family Capacities Family Capacities 77. Accept responsibility for actions 3, 4 Independence Responsibilities Independence Relational Mat. 83. supporting parents financially 1 Independence Responsibilities N/A N/A Note: For the following Role Trans., Role Transition; Relational Mat., Relational Maturity; Norm Comp., Norm Compliance; N/A, not applicable because not included in the survey items. aWhere two factors are listed, the variable yielded factor loadings greater than .30 on more than one factor.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 29

Discussion

In this document we have examined and described two approaches toward the scale

construction of importance and attainment of emerging adulthood using 40 items derived from

the literature on emerging adulthood. This document has described the sample characteristics and

items used to construct interpretable and reliable measures of emerging adulthood. The measures

are also consistent with other analyses using the MoA items.

We had taken a two approach toward scale construction from the 40 MoA items in the

EAMMI survey. Despite differences in approaches, empirical comparisons of comparisons of

deductive (theory-derived) and inductive (sample-based) constructs revealed empirical

similarities. Both approaches derive eight subscales to the 40 items, both cover similar

conceptual domains, both approaches correlate with each other highly, and both measures reflect

the state of the literature using these items.

There are few lessons to be taken from this analysis. First, the present analysis outlines how divergent approaches toward scale construction may lead to different, yet broadly, similar findings. We see two approaches toward understanding the varieties of emerging adulthood(s)

(Arnett, 2011), one approach that closely aligns with the structure of the 20 importance items in the EAMMI survey, and another, deductive approach that reflects the genesis of the emerging adulthood construct as from theoretical distinctions based on prior literature and pilot studies.

Second, the dualistic approaches we have described underscore the interlocked nature of

methods and theory. Despite divergent approaches, both J. Grahe and N. Fosse relied on theory,

interpretation, and on dissemination of findings with collaborators. The process of scale

construction outlined in this document reveal connections between method and of theory,

regardless one’s “style” of scientific reasoning (Hacking, 1994).

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 30

Third, results reveal that clustering techniques, whether such methods lump latent

constructs or latent classes, can be conducted quite easily, yet pose numerous problems on the

types of items that intended to capture the breadth of experiences in emerging adulthood. In

particular, cluster analytic techniques work better for ordinal scales such as the importance items, yet, they are potentially misleading when they are applied to binary items or toward the cluster analysis of heterogeneously constructed Likert-items. In most applications of factor analyses and of evaluation of internal reliability, there are sometimes no easy solutions nor any ‘stopping rule’ in determining the number of latent variables or latent classes (Nylund, Asparouhov, & Muthén,

2007).

While a theoretical, deductive approach appear warranted, the present analysis reveals the need to maintain theoretical clarity when assessing and conceptualizing socioemotional markers of adulthood. In contrast, risky behavior and role transitions, already prevalent in numerous surveys and in the literature, appear to be more clearly defined; in some cases, it may be more beneficial to use one carefully-selected single-item measure than an ambiguous umbrella construct. More theory development and new subscales are needed to clarify and distinguish features of emerging adulthood. Some measures, such as the Inventory of the Dimensions of

Emerging Adulthood (IDEA) inventory combine conceptual clarity, theoretical relevance, and high reliability (Reifman, Arnett, & Colwell, 2007). (For details of these and other constructs

EAMMI study, see the OSF project main page: https://osf.io/yjdaf/).

Finally, this analysis reveals the need to triangulate construct creation in emerging

adulthood research. New horizons of emerging adulthood research require examining its cultural

variation across social groups (e.g., race/ethnicity, gender, socioeconomic status) and across

nation-states Arnett (Arnett, 2012). We find that triangulating scale ‘styles’ of scientific thought

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 31 can provide insights into not only the social patterning of emerging adulthood, but also into the nature of the structure of the object of inquiry itself. With increasing globalization through social media and increased economic exchange, youth identities have become more globalized (Jensen

& Arnett, 2012), leading to potentially new dimensions of emerging adulthood that remain to be conceptualized and analyzed. In addition to the varieties of emerging adulthood(s) that warrant further investigation (Arnett, 2011), emergent forms may demand novel measures and analyses we have yet to consider (Jensen, Arnett, & McKenzie, 2011).

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 32

References

Arnett, J. J. (1997). Young people’s conceptions of the transition to adulthood. Youth & Society,

29(1), 3–23.

Arnett, J. J. (2001). Conceptions of the Transition to Adulthood: Perspectives From

Through Midlife. Journal of Development, 8(2), 133–143.

doi:10.1023/A:1026450103225

Arnett, J. J. (2003). Conceptions of the Transition to Adulthood Among Emerging in

American Ethnic Groups. New Directions for Child and Adolescent Development,

2003(100), 63–76. doi:10.1002/cd.75

Arnett, J. J. (2011). Emerging adulthood(s): The of a new life stage. In L.

Jensen Arnett (Ed.), Bridging cultural and developmental approaches to psychology:

New syntheses in theory, research, and policy (pp. 255–275). New York, NY: Oxford

University Press.

Arnett, J. J. (2012). New Horizons in Research on Emerging and Young Adulthood. In A. Booth,

S. L. Brown, N. S. Landale, W. D. Manning, & S. M. McHale (Eds.), Early Adulthood in

a Family Context (pp. 231–244). Springer New York. Retrieved from

http://link.springer.com/chapter/10.1007/978-1-4614-1436-0_15

Badger, S., Nelson, L. J., & Barry, C. M. (2006). of the transition to adulthood

among Chinese and American emerging adults. International Journal of Behavioral

Development, 30(1), 84–93. doi:10.1177/0165025406062128

Balanda, K. P., & Macgillivray, H. L. (1988). Kurtosis: A Critical Review. The American

Statistician, 42(2), 111–119. doi:10.1080/00031305.1988.10475539

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 33

Barry, C. M., & Nelson, L. J. (2005). The role of religion in the transition to adulthood for young

emerging adults. Journal of Youth and Adolescence, 34(3), 245–255.

Bland, J. M., & Altman, D. G. (1997). Statistics notes: Cronbach’s alpha. British Medical

Journal, 314(7080), 572.

Burisch, M. (1984). Approaches to inventory construction: A comparison of merits.

American , 39(3), 214–227. doi:10.1037/0003-066X.39.3.214

DeVellis, R. F. (2011). Scale development: Theory and applications (Vol. 26). Thousand Oaks,

CA: Sage Publications.

Gelman, A. (2004). Bayesian data analysis. Boca Raton, Fla.: Chapman & Hall/CRC.

Green, S. B., & Yang, Y. (2008). Commentary on Coefficient Alpha: A Cautionary Tale.

Psychometrika, 74(1), 121–135. doi:10.1007/s11336-008-9098-4

Hacking, I. (1994). Styles of Scientific Thinking or Reasoning: A New Analytical Tool for

Historians and Philosophers of the Sciences. In K. Gavroglu, J. Christianidis, & E.

Nicolaidis (Eds.), Trends in the Historiography of Science (pp. 31–48). Springer

Netherlands. Retrieved from

http://link.springer.com.ezproxy.library.tufts.edu/chapter/10.1007/978-94-017-3596-4_3

Jensen, L., & Arnett, J. J. (2012). Going Global: New Pathways for Adolescents and Emerging

Adults in a Changing World. Journal of Social Issues, 68(3), 473–492.

doi:10.1111/j.1540-4560.2012.01759.x

Jensen, L., Arnett, J. J., & McKenzie, J. (2011). Globalization and Cultural Identity. In S. J.

Schwartz, K. Luyckx, & V. L. Vignoles (Eds.), Handbook of Identity Theory and

Research (pp. 285–301). Springer New York. Retrieved from

http://link.springer.com/chapter/10.1007/978-1-4419-7988-9_13

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 34

Jolliffe, I. (2014). Principal Component Analysis. In Wiley StatsRef: Statistics Reference Online.

John Wiley & Sons, Ltd. Retrieved from http://onlinelibrary.wiley.com.ezp-

prod1.hul.harvard.edu/doi/10.1002/9781118445112.stat06472/abstract

Kaiser, H. F. (1974). An index of factorial simplicity. Psychometrika, 39(1), 31–36.

Kotrlik, J., Williams, H., & Jabor, K. (2011). Reporting and Interpreting Effect Size in

Quantitative Agricultural Education Research. Journal of Agricultural Education, 52(1),

132–142. doi:10.5032/jae.2011.01132

Nelson, L. J., Padilla-Walker, L. M., Carroll, J. S., Madsen, S. D., Barry, C. M., & Badger, S.

(2007). “If you want me to treat you like an adult, start acting like one!” Comparing the

criteria that emerging adults and their parents have for adulthood. Journal of Family

Psychology, 21(4), 665–674. doi:10.1037/0893-3200.21.4.665

Nezlek, J. B. (2011). Multilevel Modeling for Social and . Los Angeles,

CA: Sage Publications.

Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the number of classes in

latent class analysis and growth mixture modeling: A Monte Carlo simulation study.

Structural Equation Modeling, 14(4), 535–569.

Reifman, A., Arnett, J. J., & Colwell, M. J. (2007). Emerging adulthood: Theory, assessment,

and application. Journal of Youth Development, 2(1), 1–12.

Schoon, I. (2015). Diverse Pathways: Rethinking the Transition to Adulthood. In P. R. Amato,

A. Booth, S. M. McHale, & J. V. Hook (Eds.), Families in an Era of Increasing

Inequality (pp. 115–136). Springer International Publishing. Retrieved from

http://link.springer.com/chapter/10.1007/978-3-319-08308-7_9

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 35

Shanahan, M. J. (2000). Pathways to Adulthood in Changing Societies: Variability and

Mechanisms in Life Course Perspective. Annual Review of Sociology, 26(1), 667–692.

doi:10.1146/annurev.soc.26.1.667

Sijtsma, K. (2009). On the use, the misuse, and the very limited usefulness of Cronbach’s alpha.

Psychometrika, 74(1), 107–120.

Smith, G. T., Fischer, S., & Fister, S. M. (2003). Incremental Validity Principles in Test

Construction. Psychological Assessment, 15(4), 467–477. doi:10.1037/1040-

3590.15.4.467

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 36

Appendix.

Table 1.

Mean and Standard Deviation of Importance Items in EAMMI Survey

Item n M SD 67. Married 1336 2.89 1.14 68 Have at least one child 1334 3.13 1.13 73 Long-term relationship 1334 2.62 1.13 69 Long term Career 1336 2.37 1.06 83 Capable Support Parents 1335 2.57 0.99 76 Capable Caring Children 1335 1.84 0.96 66 Finished Education 1335 2.21 1.05 75 Capable Supporting Family 1337 1.82 0.94 71 Avoid Illegal Drugs 1335 2 1.14 70 Avoid Becoming Drunk 1337 2.59 1.1 79 Avoid Drunk Driving 1336 1.45 0.84 72 Use Contraception 1332 1.91 1.04 80 Interact with Parents as Equals 1334 1.69 0.8 82 Become Less self-oriented 1334 1.8 0.78 81 Control Your Emotions 1335 1.96 0.83 74 Make Independent Decisions 1335 1.31 0.61 77 Accept Responsibility for Actions 1335 1.18 0.53 64 Financially Independent 1334 1.54 0.7 65 No Longer Live with Parents 1335 1.75 0.86 78. Be Employed Full Time 1333 1.81 0.88

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 37

Table 2.

Tests of Sampling Adequacy of 20 Importance Items

Item KMOa SMCb 67. Married .84 .69 68 Have at least one child .83 .66 73 Long-term relationship .93 .54 69 Long term Career .92 .47 83 Capable Support Parents .93 .32 76 Capable Caring Children .84 .54 66 Finished Education .92 .34 75 Capable Supporting Family .84 .57 71 Avoid Illegal Drugs .81 .50 70 Avoid Becoming Drunk .82 .42 79 Avoid Drunk Driving .85 .33 72 Use Contraception .88 .23 80 Interact with Parents as Equals .90 .21 82 Become Less self-oriented .88 .26 81 Control Your Emotions .88 .27 74 Make Independent Decisions .77 .27 77 Accept Responsibility for Actions .77 .28 64 Financially Independent .85 .21 65 No Longer Live with Parents .85 .16 78. Be Employed Full Time .92 .34 Note. Tests of adequacy assess degree of inter-item covariance. a. Kaiser-Meyer-Olkin measure. b. Squared multiple correlations (SMC) of variable against all others (R2).

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 38

Table 3.

Unrotated Principal Components of 20 Importance Items

Component Eigenvalue Diff.a Percentb Cumul.c 1 5.59 3.25 28% 28% 2 2.35 0.92 12% 40% 3 1.43 0.12 7% 47% 4 1.31 0.34 7% 53% 5 0.97 0.09 5% 58% 6 0.88 0.05 4% 63% 7 0.83 0.07 4% 67% 8 0.77 0.07 4% 71% 9 0.69 0.04 3% 74% 10 0.65 0.05 3% 77% 11 0.60 0.01 3% 80% 12 0.59 0.01 3% 83% 13 0.58 0.01 3% 86% 14 0.56 0.01 3% 89% 15 0.55 0.11 3% 92% 16 0.45 0.08 2% 94% 17 0.37 0.03 2% 96% 18 0.34 0.07 2% 98% 19 0.27 0.07 1% 99% 20 0.20 N/A 1% 100% Note. Principal components analysis based on 20 importance items, Stata/SE 13.0. Retained components present in bold. Total number of eigenvalues equal the number of items used to estimate total item variance. a. Difference between the row component eigenvalue and the proximate eigenvalue. b. Proportion of variance explained by each eigenvalue. c. The cumulative proportion of variance explained by the eigenvalues.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/) Markers of Adulthood Subscale Development 39

Table 4.

Item Loadings from Varimax Rotation of Four Factors, Replication in Sata/SE

Item Component Unexplained 1 2 3 4 67. Married 0.43 -0.02 -0.02 0.01 0.28 68 Have at least one child 0.45 0.00 -0.09 -0.01 0.29 73 Long-term relationship 0.35 0.05 0.02 0.07 0.38 69 Long term Career 0.34 -0.07 0.15 -0.01 0.42 83 Capable Support Parents 0.30 0.16 -0.09 0.00 0.56 76 Capable Caring Children 0.23 0.16 0.20 -0.07 0.49 66 Finished Education 0.24 -0.05 0.21 0.03 0.58 75 Capable Supporting Family 0.20 0.10 0.32 -0.08 0.42 71 Avoid Illegal Drugs 0.02 -0.04 0.03 0.56 0.29 70 Avoid Becoming Drunk 0.07 -0.02 -0.07 0.52 0.39 79 Avoid Drunk Driving -0.13 0.14 0.07 0.43 0.46 72 Use Contraception -0.02 -0.02 0.03 0.44 0.58 80 Interact with Parents as Equals 0.05 0.48 -0.05 -0.08 0.52 82 Become Less self-oriented 0.04 0.46 -0.15 0.10 0.49 81 Control Your Emotions 0.09 0.42 -0.12 0.04 0.55 74 Make Independent Decisions -0.18 0.37 0.26 -0.08 0.49 77 Accept Responsibility for Actions -0.22 0.36 0.23 0.00 0.48 64 Financially Independent -0.04 -0.08 0.50 0.00 0.52 65 No Longer Live with Parents -0.06 -0.08 0.45 0.03 0.61 78. Be Employed Full Time 0.08 0.01 0.38 0.07 0.52 Note.Varimax rotation of 20 importance items. Selected items for mean score construction bold. Analyses run in Stata/SE based on the correlation matrix of items. n = 1309. Rotation conduced after orthogonal rotation of all 20 items, retaining first four components. Unexplained column on the far right provides a measure of unexplained variance of each item. Retaining 4 components results in loadings that can a most explain 53% of the variance in the items.

Research Report for EAMMI Survey Project, Open Science Framework (https://osf.io/p8nwq/)