High Commitment Management in the U.K.: Evidence from the Workplace Industrial Relations Survey, and...

High Commitment Management in the U.K.:

Evidence from the Workplace Industrial

Relations Survey, and Employers’ Manpowerand Skills Practices Survey

Stephen Wood1,3 and Lilian de Menezes2

Are the practice s widely associated with the high commitment or involvement

model, such as job flexibility and minimal status differences, actually used in

conjunction with e ach othe r? O r rathe r are the y be ing u se d, as some

commentators speculate, in a fragmented or ad hoc manner? The authors use

Late nt Variable Analysis to asse ss whether practice s identified with high

commitment managemen t do form a unity. They are simultaneously attempting

to see if such practice s can be used as indicators for measuring an underlying

high commitment orientation on the part of management . The analysis uses data

from the 1990 UK Workplace Industrial Relations Survey and its sister survey,

the Employers’ Manpower and Skills Practices Survey, on the use of a range of

high commitme nt practices across the whole economy. The evidence sugge sts

that there is an identifiable pattern to the use of high commitment practice s.

Four progre ssive style s of high commitme nt manage me nt (HCM) we re

discovered. Though the use of it in its entirety is still relatively rare in the UK,

the proportion of organizations with me dium le ve ls of high commitme nt

management is higher than is perhaps commonly assumed. High degrees of high

commitment management are not necessarily associated with nonunion workplaces.

The research also demonstrates that HCM does have some performance effects,

though they are not unique to it since those organizations that adopt high

commitment management in its entirety do not perform better on any performance

criteria than all others, but they do perform better than some types.

K EY WORDS: high commitme nt manage me nt; trade unionism; human

resources manage ment and organizational performance ; latent variable analysis.

Hum an Relations, Vol. 51, No. 4, 1998

485

0018-7267/98/0400-0485 $15.00/1 Ó 1998 The Tavistock Institute

1Methodology Institute and Departme nt of Industrial Relations, The London School of Eco-nomics and Political Science, London WC2A 2AE, England.

2Department of Mathematical and Computing Sciences, Goldsmiths College, Unive rsity ofLondon, Ne w Cross, London SE14 6NW, England.

3Requests for reprints should be addressed to Stephen Wood, Methodology Institute and De-partment of Industrial Relations, The London School of Economics and Political Science,

Houghton Street, London WC2A 2AE, England.

INTRODUCTION

A new form of organization which has variously been labe led the high

commitment, high involve ment, high performance mode l is central to much

of the discussion in organization theory and employee re lations (Appel-

baum & Batt, 1994; Beaumont, 1995; Boxall, 1992; Fernie & Metcalf, 1995;

Guest, 1995; Kalle berg et al., 1996, Chap. 6; Kochan & Osterman, 1994;

Lawle r, 1986; Lawle r et al., 1995; Osterman, 1994; Store y, 1992, 1995;

Wood & Albanese, 1995). For some, it is unive rsally applicable , while for

others, it is one among several mode ls of good personnel manage ment, its

relevance be ing continge nt on the organization ’s context. Despite such dif-

ferences, the commitment mode l, particularly popularize d by Walton (1985)

is wide ly portraye d as be ing especially suited to fluid situations. It arrived

at the forefront of discussion in the late 1980s precise ly because it appe ared

to be critical for the technological change s and continuous improve ment

which were seen as increasingly required in all organizations as they face

highly competitive and unstable product marke ts.

High commitment management (HCM) is characterized by the use of

such personne l practice s as information disse mination, proble m-solving

groups, minimal status diffe rences, job flexibility, and teamworking; but

more importantly, so its archite cts would argue , it entails using them in

combination. Moreover, unde rlying the ir use is assumed to be a commit-

ment on the part of employe rs to their employe es base d on an underlying

conception of them as asse ts or resources to be developed rathe r than as

disposable factors of production. A key initial que stion is then whe ther

there is any patte rn to the use of these high commitment practices. Having

identified a patte rn, we can then answe r the questions which lie at the

center of current debate about HCM, namely:

(1) is it more prevalent in nonunion organizations and do trade un-

ions in fact constrain the introduction of high commitment man-

agement,

(2) do those organizations which adopt it actually outperform all oth-

ers?

Alternative ly, if there is no patte rn, we would have no choice but to limit

analysis to the incidence and impact of individual high commitment prac-

tices (as for example Fernie et al., 1994, have done).

The results of previous analyse s of high commitment practice s in the

U.K. imply that we should not expect a patte rn in the ir usage . The initial

analysis of the U.K. Workplace Industrial Relations Survey (Sisson, 1993,

1995, p. 106; Millward, 1994, p. 129) discovered fragments of HCM and has

been especially influential in portraying the adoption of high commitment

practices as a piecemeal affair. This was largely inferred from an appare nt

486 Wood and de Menezes

low adoption of individual high commitment practice s and consequently even

smalle r use of the whole package of high commitment practice s. Moreove r

this pick-and-mix approach to the use of such practice s was what many ex-

pected, as it conformed to their view that, because of an overemphasis on

short-te rm and/or financial measure s of pe rformance, management in Britain

approache s personne l matters in an ad hoc, or pragmatic manner. Some have

also implie d that this ad hocism lies behind the seeming co-existence of un-

ionism and high commitment practice s and the lack of any strong perform-

ance effects resulting from their use (Edwards, 1995) .

It is possible that this limited discove ry of HCM in the 1990 Workplace

Industrial Relations Surve y (WIRS90)4 is, however, an artifact of the type of

analysis (univariate ) conducted so far, or even of the research instrument it-

self. For the relative low usage of practices associated with HCM cannot, in

and of itse lf, be take n as indicative of ad hocism. Even if only 20% of organi-

zations use a particular HCM practice , it may be that the same 20% also use

another practice. Hence we may be able to identify a subse t of high commit-

ment practices which are used in conjunction with each other and conse-

quently infe r that some high commitment management is being practiced.

This paper reports research, using the WIRS90 data, based on a detailed

analysis of the associations between a comprehensive set of HCM practice s.

It focuse s on the question of whether high commitment practices are used

in concert with each othe r. It differs from the earlie r analyse s of WIRS since

it uses multivariate statistical methods to examine the relations among the

high commitment practices. In addition, it supplements the WIRS90 dataset

with information from its sister survey, the Employe rs’ Manpower and Skills

Practices Survey. Having establishe d that there are high commitment ap-

proache s, it then examine s: (a) whether high commitment management is

antithe tical to unionism, and (b) whether those organizations which adopt it

perform better than othe rs. The paper ope ns with an account of the study,

the research design, the dataset, the measures of high commitment practice s,

and the statistical models used to assess the relationship between them.

THE STUDY

The Core Concept: High Commitment Management

High commitment manage ment is assumed to be aimed at eliciting a

strong commitment to the organization, so that behavior is primarily self-regu-

High Commitmen t Managem ent in the U.K. 487

4Although the third survey in the Workplace Industrial Relations Survey serie s was termed

the Workplace Employe e Relations Survey, it tends to still be known as WIRS, usuallyWIRS3 or WIRS90. In what follows, we will mainly use WIRS90 to refer to the dataset

which we are using, and only WIRS when we are specifically referring to the series of surveys.

lated rather than controlled by sanctions and pressures external to the individ-

ual, and consequently relations within the organization are based on high trust

(Wood & Albane se, 1995) . The central assumption is that managements

through designing high commitment work systems are creating the conditions

for employees to become highly involve d in the organization and identify with

its overall goals. In this way it is then assumed that they will perform consis-

tently at a high level, as well as show initiative and a willingness to “put them-

selves out for the organization.” Following the high commitment route entails

adopting specific practices, including recruitment practices which aim to attract

and select highly committed and flexible people, internal labor markets which

reward commitment and training with promotion and job security; and methods

of direct communication and teamworking. They may be summarized thus:

· the development of career ladders, and an emphasis on trainability

and commitment as highly valued characte ristics of employees at all

leve ls of the organization;

· a high level of functional flexibility with the abandonme nt of pote n-

tially rigid job descriptions;

· the reduction of hie rarchies and the ending of status diffe rentials at

least between white-collar and manual or blue -collar workers, if not

between managers and workers;

· a heavy reliance on the team structure for disseminating information

(teambriefing) , structuring work (teamworking) , and problem solving

(quality circles).

Some writers also associate other management practice s with HCM. Guest

(1987, p. 515) , Lawler (1986) , and Walton (1985) , for example , include job

redesign aimed at the full utilization of human resources in their set of

high commitment practice s. A full list of practices which are sometimes,

though not invariably, associated with high commitment manage ment

would include :

· job (re)design being something which manage ment consciously does

in order to provide jobs that have a conside rable leve l of intrinsic

satisfaction;

· a policy of no compulsory lay-offs or redundancie s and permanent

employme nt guarante es with the possible use of temporary workers

to cushion fluctuations in the demand for labor;

· new forms of assessment and payment systems, more specifically

merit pay and profit-sharing;

· a high involve ment of employe es in the manage ment of quality;

· the provision of extra-we lfare facilitie s;

· the sharing of information about the organization and encourage -

ment of employees to identify with its overall goals and future.


Research Design

The obje ctive of this research is to see, first, if there is any patte rn to

the use of high commitment practices, and specifically to identify if they

are used in conjunction with each othe r. Second, to assess whether we can

use a set of high commitment practice s as indicators of some underlying

approach or even commitment on the part of manage ment toward labor.

So, rather than measuring management’s orientations directly, we will be

using a set of more readily obse rvable institutionalize d arrange ments in our

attempt to measure high commitment management.

Latent variable analysis (as described in the Appendix) is an appropriate

statistical method for tackling the twin objectives of measuring high commit-

ment manage ment and seeing whether practices commonly associate d with it

form a unity. It uses a set of indicators such as responses to questions in a

survey in order to identify and deve lop scales of underlying abilitie s, orienta-

tions or values. It assesses whether any association which might exist between

a set of items can be explaine d by a common factor. Here we assume that

data on whether manage ments adopt a high commitment approach and the

underlying orientation toward labor associate d with it are not readily accessi-

ble , but observations of the high commitment practices are available . We thus

use latent variable models to see if an underlying commitment approach to

employees explains any association which may exist between a set of practices

associate d with high commitment.

In this study, the indicator variable s, the use of high commitment prac-

tices, are binary, and therefore we use two models that can be seen as part of

the family of factor analysis, namely:

· a late nt trait m ode l, the ‘logit-probit m ode l’, de ve lope d by

Bartholome w (1987) , in which the late nt variable (underlying factor)

is continuous as it is in factor analysis, but the indicator variable s

are binary;

· the ‘latent class mode l’ (Lazarsfe ld & Henry, 1968) , where both in-

dicator and late nt variable s are categorical.

By using latent class analysis homoge neous groups or individuals, in our

case workplace s, can be identifie d. These groups are , in contrast to cluster

analysis, based on the position of the establishme nts in the late nt scale ,

that is, on the level of high commitment management in the workplace .

Details of these mode ls are given in the Appendix, and information on the

software—developed at the LSE—is available directly from the authors.

If a model does fit the data, rather than simply aggregating the number

of high commitment practices used in a particular workplace , we can use

the resulting latent variable scores as a measure of high commitment man-

agement. This can be used in further analysis either as a depende nt variable


in a regression mode l aimed at assessing its relationship to unionism or

other variable s, or alternative ly as an independe nt variable when analyzing

variability in organizational performance.

The Datasets

The Workplace Industrial Relations surve y (WIRS) is aimed at achiev-

ing a “better understanding of the processes which unde rlie employme nt

relationships” through a large-scale surve y of a “broad range of industrial

relations and employme nt practices” (foreword to Millward et al., 1992) .

It is based on a representative sample of the whole economy, though it is

limite d to workplace s with 25 or more employees. Two thousand and sixty-

one establishme nts—which reflected a response rate of 82.7% (Millward

et al., 1992, p. 376)—were include d in the 1990 survey. Of these 30% are

in the public sector. The inte rviewing took place between February and

October 1990. WIRS’s unit of analysis is the establishme nt (e.g., a factory

in a large multisite manufacture r) and it concentrates on workplace (rather

than companywide ) matters. The core questions are particularly concerned

with industrial relations practice s, including communication methods and

pay determination procedures, industrial conflict, and certain employme nt

practice s such as fixe d term contracts.

The Employe rs’ Manpowe r and Skills Practice s Surve y (EMSPS) is

based on intervie ws in 1693 (a response rate of 84% ) of the establishme nts

surveyed in WIRS90. The data colle ction took place immediately following

the fieldwork for WIRS90, between November 1990 and October 1991. The

survey’s objective was “to explore employe rs’ skill formation practice s, includ-

ing their skill needs, recruitment practice s, training and other employee-re-

lated activitie s” (Employers’ Manpowe r and Skills Practice s Survey, 1994).

Analyse s of some of the key dimensions of the surve y, e.g., the size of the

establishme nt and the extent of unionization, as well as the use of particular

HCPs, showed that the EMSPS subsample remained representative .

The two surve ys toge ther provide information on a wide range of

HCPs, the main areas on which information is lacking be ing teamworking

and job guarante es. Within WIRS90 there are que stions on the following

practice s: quality circles/problem-solving groups, appraisal, me rit pay,

profit-sharing, welfare facilitie s or fringe benefits, information disclosure ,

teambriefing, top manage ment briefing, suggestion schemes, and attitude

surve ys; while EMSPS contains information on the following practice s: in-

ternal recruitment, multiskilling, social skills as a selection criterion, team-

working skills as a selection criterion, and training needs analysis.

There are no direct que stions about the extent of hierarchy in either

surve y, but re sponde nts are aske d whe the r manual workers are paid


monthly (monthly pay) , receive payment through a bank account (cashle ss

pay) , or do not have to clock-in (no clocking-in) . These can be used as

indicators of whether they and othe r lower-level participants in organiza-

tions are being treated in the typical way that white-collar and management

staff have in the past.

There are in WIRS90 que stions relating to functional flexibility and

employee involve ment. But both ask whether during the last 3 years man-

agement has introduce d any changes which increase the flexibility of the

workforce or are aimed at increasing employee involve ment. Neither ques-

tion yie lds a measure of either functional flexibility or involve ment, since

organizations which indicate d that they have made no change in the past

3 years may consist of two type s: (a) those which had a low leve l of flexi-

bility or involve ment 3 years ago and have made no attempt to increase it

in the past 3 years, and (b) those which had a high leve l of flexibility or

involve ment both then and when surve yed.

Indicators Used in the Study

Quality Circles/Problem -Solving G roups

This is based on a question in which responde nts were asked if man-

agement used “regular meetings among work-groups or teams at least once

a month to discuss aspe cts of the ir performance , such as quality circle s and

other problem-solving groups.”

Teambriefing

This is based on a que stion which asked (responde nts) if in their work-

place management used “regular meetings (at least once a month) between

junior managers/supe rvisors and all the workers for whom they are respon-

sible —these are sometimes known as brie fing groups or teambriefing—to

communicate or consult with their employees.”

Top Managem ent Briefing

This measure s whether the establishme nt has regular meetings (at least

once a month) between senior manage rs and all sections of the workforce

(either altoge ther or section by section) .


Monthly Pay

This indicate s whe the r all e mploye e s in the workplace are paid

monthly.

Cashless Pay

This measure s whether all employe es receive their payme nt through

a bank account.

No Clockin g in

This measure s if all employe es do not have to clock in.

Internal Recruitment

This is based on a question asking whether a preference exists for in-

ternal recruitment.

Multiskillin g

Responde nts were asked if “any employees at this establishme nt have

been traine d to be multiskille d.”

Social Skills as a Selection Criterion

Responde nts were aske d to pick from a range of 18 qualitie s the five

most important ones which they look for when recruiting for the two large st

occupational groups. One such quality is inte rpersonal skills/communica tion

skills; and an establishme nt was coded as using social skills if this was seen

as an important criterion for one or both of the two occupational groups

unde r scrutiny.

Teamworking Skills as a Selection Criterion

This measure was constructed in the identical way to the social skills

one , but is base d on the use of an individual ’s “ability to get on with others

in a team/fitting-in” as a selection criterion.


Train ing Needs Analysis

This is measured by whether (or not) responde nts answe red positive ly

to, “Are the training and deve lopme nt needs of employe es assessed.”

Appraisal

Respondents were directed that performance appraisal is meant to refer

to an “individual written assessment produce d periodically by manageme nt

or supervisors.” They were initially asked to provide information for a group

of employees, namely the large st among either manual or clerical/se cretar-

ial/adm inistrative . Responde nts were then aske d whether the appraisal

scheme is used for all employe es or only some. In this study, the measure

used is based on the establishme nt having appraisal for all employees.

Merit Pay

Responde nts were asked whether any employe es in eight occupational

groupings “receive merit pay or pay related to the assessment of individual

performance?” The eight groups range d from unskille d manual workers to

middle /senior managers. In this study, as we focus on nonmanage rial em-

ployees, the measure concentrate d on the use of merit pay for the following

occupational groupings: unskille d manual workers, se miskille d manual

workers, skilled manual workers, clerical/adm inistrative /secretarial, supe rvi-

sors/foremen, junior technical, or professional. The merit pay item used in

this study me asure d whether it was used for all these groups, though a

second measure based on when it was used for some, that is, for at least

one of the six occupational groups, was used in some secondary analysis.

Profit-Sharing

Workplace s are taken as having profit-sharing if they have one or other

of what are called in the WIRS90 questionnaire : (a) profit-re lated payments

or bonuses (including those covered by the 1987 Finance Act), or (b) a

deferred profit-sharing scheme—where profits are put in a trust fund which

acquire s shares in the employing company for employees. This practice is

confined to the private sector.


Employee Share Option s

This is base d on whether respondents claim to have a share ownership

scheme, and again is only applicable to the private sector.

Welfare Facilities or Fringe Benefits

Responde nts were asked whether the establishme nt provide d (a) free

or subsidize d food or meals, (b) sick pay over and above statutory require-

ments (hereafter sick pay), (c) an occupational pension scheme, and (d) a

standard working week of less than 36 hours (hereafter referred to as a

shorte r working week) for both all or some employe es. The variable s used

in this research are based on whether the benefits were provide d for all

employees.

Information Disclosure

Interviewees were asked about the type of information disclose d. Four

of the most important items on which information is given to employe es

or the ir representative s were used in this study, namely (a) labor produc-

tivity, (b) the level of wage /salary costs, (c) internal inve stment plans, (d)

the financial position of the establishme nt.

In some studies of human resource practice s (e.g., Osterman, 1994)

intraestablishme nt differences are measured in terms of the proportion of

the workforce involve d in a particular practice . In this study, most practices

are conceived to extend across the whole plant, e .g., information disclosure ,

or to be only relevant as indicators of HCM if and only if they apply to

the whole workforce .5 The exception to this was multiskilling, since the

question simply asked if e mploye es at the establishme nt have had mul-

tiskilling training, and thus the proportion of the workforce affected may

vary between establishme nts.


5For many items the survey question specifically aske d whether the practice was used for allemployees, e.g., cashless pay and teambriefing, while in other cases its widespread use (as

in training needs analysis) can be inferred from a positive answer. In the case of qualitycircles/problem-solving groups, variation in usage between departme nts, within estab-

lishments, is possible at least over time. Nonetheless, the question asked whether they wereused “as a matter of policy” and it is likely that respondents would have only reported using

such groups when these were a core part of their approach to employee relations as wasfound to be the case in follow-up interviews in firms that had a participated in a mail survey

conducted by one of the authors (see Wood, 1996) .

THE ANALYSIS OF HIGH COMMITMENT MANAGEMENT

The Use of High Commitment Practices

In the latent variable analysis, we used unwe ighted frequencie s, i.e .,

the actual numbe r of workplace s which adopt a particular practice . We did

not need to correct for sample bias, since the latent variable models which

we used assume that the interrelationships (cross-corre lation structure ) are

independent of the sample structure. In othe r words, the underlying pattern

is considered to be the same within groups, regardle ss of the sample design.

From Table I, we can see that there was a sizeable proportion of plants

using the set of high commitment practices which are associate d with HCM,

the exception being merit pay. The use of extra-welfare facilitie s, cashless

pay, and training need analysis was very widespread. But overall, the per-

centage of workplace s which used a particular practice was normally below

Table I. High Commitment Practices: Frequency of Usage in the Samplea

Full

sample

Private

sector

Public

sector

Internal recruitment 53.3 56.2+ 42.1

Multiskilling 56.7 58.8+ 48.4Human re lations skills as a selection criterion 83.4 81.9 89.4+

Social skills 57.2 55.6 63.3Team working skills 64.9 63.2 71.3

Training needs analysis 86.3 86.2 86.8Direct communication 76.2 74.9 80.8

Quality circles 37.3 35.9 43.0+

Teambriefing 56.0 54.0 63.9+

Top management briefing 47.6 46.9 50.1Monthly pay 33.3 33.9 30.9

Cashless pay 71.6 69.8 78.5+

No clocking in 17.9 17.6 19.2

Appraisal 30.8 33.5+ 20.3Merit pay 1.4 1.5 0.9

Financial involvement 53.6 67.2 ¯Profit-sharing 37.4 46.8 ¯Employee share option 38.7 48.8 ¯

Welfare facilities 77.0 79.4 67.8

Food 50.7 52.8 42.7Sick 69.3 72.8+ 55.9

Pension 73.1 75.0+ 65.6Shorter working week 17.9 19.2 17.6

Information disclosure 80.8 82.5 74.2Labor productivity 29.9 32.1+ 21.8

Salary costs 30.3 32.2+ 35.2Internal investment 51.5 55.8+ 67.0

Financial position of the establishment 69.2 69.8 67.0

aSample sizes: full = 1690, private = 1341, public = 349. + Significantly more likely to have

practice s (at or below 1% leve l).


50% . Exactly 50% of the establishme nts used between six and ten of the

HCPs include d in the study. Only 5% used above 14 and 25% had five or

less practices. The two modes were seven and eight items (practice s), and

227 (out of a sample of 1441) establishme nts had these totals.

As one might have expe cted, Table I shows that the usage within the

full sample is not very different from the private sector. Some items such

as inte rnal recruitment are, however, significantly more like ly to be used

in the private sector than in the public sector, while othe rs such as team-

brie fing are more like ly to be found in the public sector. The analysis that

we report here will conce ntrate on the whole sample , but further work

could usefully explore the potential diffe rences between sectors.

Laten t Variab le Analysis

The obje ctive of the late nt variable analysis was first to assess whether

there is some common factor underlying the data, and second to ascertain

whether high commitment management is a well-de fined continuous vari-

able (a latent trait mode l fits the data) or rather a set of progre ssive styles

of management (a late nt class model is preferred). We shall report the

main stages of the analysis.

Constructing the Indicators

Preliminary analysis of the 23 high commitment practices on which we

have information (as in Table I) revealed that certain subsets of practices

were independe ntly re lated to each other, name ly: the welfare practice s

(food, extra sick pay, pension, shorter working week), skill formation and

selection criteria (multiskilling, internal recruitment, training needs analysis,

social and team skills) ; direct communication methods (quality circle s,

teambriefing, top management briefing) ; and the information disclosure (on

labor productivity, salary costs, internal inve stment, financial position of the

establishme nt) items. The information disclosure on salary costs appe ared

to have little discriminatory power, nor did the selection criteria items.

Four composite indices were then constructe d. Each was coded 1 if a

workplace used at least one of the practices falling in it, namely: welfare;

skill formation; direct communication; and information disclosure .

Latent Trait Analysis

The initial model was based on a set of 11 indicators: skill formation,

direct communication, merit pay, appraisal, no clocking in, welfare, profit-


sharing, employe e share options, monthly pay, cashle ss pay, and information

disclosure . The comparison of observed and expected response s to pairs of

items, as described in the goodne ss-of-fit section of the Appendix, revealed

that there was a strong corre lation between profit-sharing and employe e

share options in both models. Hence, these items were excluded which also

seemed sensible since the analysis focused on the whole economy.

A one -factor model without eithe r of the financial involve ment prac-

tices was not satisfactory due to large residuals between the obse rved and

expected frequencies of some combinations of practices. Moreove r, the wel-

fare item had an extremely high loading, suggesting a threshold behavior,

that is, that the sample could merely be divide d into establishme nts which

had extra-welfare facilitie s and those which had not.

A two-factor mode l was slightly better, though the mode l was mean-

ingle ss since both factors were heavily weighte d on the welfare item, thus

reinforcing our previous conclusion that the welfare item obscures the at-

tempt to measure HCM. In addition, merit pay had very little discrimina-

tory power.

The subsequent part of the analysis concentrate d on a set of seven

items—welfare and merit pay having been exclude d: skill formation, direct

communication , appraisal, no clocking in, monthly pay, cashless pay, infor-

mation disclosure . Both models did not fit the data. Residuals between ob-

se rved and expe cted freque ncie s of pairs of items involving appraisal,

information disclosure and direct communication were large in the one-

Table II. One-Factor Model with Seven Items: Discrepancie s Betwee n Ob-served and Expe cted Responses [(O-E)2/E]a

Responses (1,1) (1,0) (0,1) (0,0)

Values > 2.5 3 3 3 2

Maximum item 5.79 18.79 15.03 41.62(3,2) (3,2) (7,2) (7,41)

Other large item 4.16 11.70 10.1 7.41(4,3) (7,2) (7,4) (3,2)

aItems: 1. skill formation, 2. direct communication, 3. appraisal, 4. no clock-

ing in, 5. monthly pay, 6. cashless pay, 7. information disclosure.

Table III. Two-Factor Model with Seven Items: Discrepancie s Betwee nObserve d and Expected Responses [(O-E)2/E]a

Responses (1,1) (1,0) (0,1) (0,0)

Values > 2.5 3 1 2 0

Maximum item 5.16 3.18 9.98(4,3) (4,3) (7,4)

aItems: 1. skill formation, 2. direct communication, 3. appraisal, 4. no clock-

ing in, 5. monthly pay, 6. cashless pay, 7. information disclosure.


factor model (Table II), while in the two-factor model a similar problem

arose with respect to appraisal and no clocking in (see Table III). Hence

we could not fit a late nt trait model to the data, that is a continuous meas-

ure of HCM was not achie ved.

Latent Class Models

In contrast, both a three- and four-class mode l fit the data. Underlying

the data, there is then a latent variable , which is categorical, and we iden-

tified four homoge neous groups of organizations. In the three-class mode l

the first class had a very low usage of HCPs (see Table IV). Nonetheless,

it had a very high probability of having skill formation, and around 50%

probability of having cashle ss pay or information disclosure . In all, 14% of

workplace s were predicted to fall in this category. The second class—54%

of workplace s—differed from the first principally on its use of direct com-

munication, and second, in its having a higher chance of using appraisal.

The third class, of which 31% of establishme nts belonge d, had a high rela-

tive usage of all items, though less than half of the workplace s in it had

appraisal or no clocking in.

The four-class model, whose parameters are shown in Table V, essen-

tially segmented the third-class from the previous (three-class) model. The

third and fourth classes were demarcated above all by their relative use of

appraisal (item 3), the former be ing characte rized by its lack of use , the latte r

by its use. The fourth also was significantly more like ly to have no clocking

in. We can thus treat the fourth class as representing a full-fle dged HCM.

Because of the high freque ncy of the skill formation variable we also

did the analysis substituting it for training needs. The results did not change

significantly. We also tested whether the mode l fit when merit pay (meas-

Table IV. A 3-Class Model (Seven Items): Prob-abilities of Having a Practice a

Class 1 Class 2 Class 3

1 0.95 0.95 0.94

2 0.17 0.86 0.853 0.05 0.35 0.36

4 0.14 0.14 0.245 0.12 0.01 1.00

6 0.57 0.61 1.007 0.49 0.89 0.89

Size 0.14 0.54 0.32

aChi-square statistic = 144 (105 df), loglikelihood ratio

statistic = 130 (105 df). Items: 1. skill formation, 2.

direct communication, 3. appraisal, 4. no clocking in,

5. monthly pay, 6. cashless pay, 7. information disclosure.


ured by whether it was used for some employees) was include d, the results,

however, were not satisfactory.

The four-class model provide d an inte rpretation which is intelligible

according to our conceptions of high commitment manage ment and im-

prove d upon the three-class mode l. Direct communication is the main

source of discrimination at the bottom end of the scale and appraisal the

main discriminator at the top. The first class could be characterized as low

HCM, the fourth as high HCM. Moreove r, there was some ordered pro-

gression from the first to the fourth. For, while the two medium classes

were clearly distinguishable by their having different subsets of practices,

the third could be seen as a progression on the second as the probabilitie s

of having many of the items was indeed greater in the third than in the

second. The third class was particularly characterized by its use of both

monthly and cashless pay which can be take n as indicators of salarie d status

or white collarization to use Koike ’s (1987) terminology.

The third group clearly represented a type of organization with a high

proportion of white -collar employees which tends to extend the degree of

involve ment and facilitie s associate d with the se employe es to all othe rs.

We cannot, however, conclude that all our classification is doing is distin-

guishing between organizatio ns dominate d by white collar and manual

work. For the highest group was not dominate d by organizations with a

high proportion of white collar employees, nor were the two groups at the

low end of the scale dominate d by organizations with a high proportion of

manual workers.

Since we fitted a latent class model to the combine d EMSPS and

WIRS data we discove red a pattern in the data which is not consiste nt

with the ad hoc or “pick-and-mix ” approach freque ntly discussed in the

lite rature. First, the identification of four classe s sugge sts that there is some

logic to the use of high commitment practice s. Second, subse ts of practices

Table V. A 4-Class Model (Seve n-Items): Probabilities of Hav-

ing a Practice a

Class 1 Class 2 Class 3 Class 4

1 0.96 0.96 0.93 0.95

2 0.26 0.89 0.78 0.923 0.09 0.38 0.00 1.00

4 0.15 0.14 0.17 0.375 0.04 0.02 0.90 1.00

6 0.51 0.60 0.98 1.007 0.53 0.88 0.87 0.91

Size 0.16 0.49 0.24 0.11

aChi-square statistic = 99 (97 df), loglikelihood ratio statistic =

106 (97 df). Items: 1. skill formaton, 2. direct communication, 3.

appraisal, 4. no clocking in, 5. monthly pay, 6. cashless pay, 7.

information disclosure.


tend to co-exist. For example , if a manage ment discloses information on

one subject, it is more like ly to disclose it on othe rs, or fringe benefits

(what we have termed we lfare items) tend to be used together and are

important in the ir own right as means of differentiating between organiza-

tions. Overall, this study has discovered distinct patte rns within the multi-

variate dimensions of the data:

· four homoge neous groups of HCM have been identifie d and hence

we have discovered types, rather than fragments, of HCM;

· welfare is a separate dimension to HCM;

· appraisal discriminate s at the top end of the scale of HCM whereas

information disclosure differentiates between organizations at the

bottom end.

The fact that we eventually excluded merit pay from our analyse s because

of its low discriminatory power cannot be taken to mean that it is definite ly

not part of HCM. It would certainly be wrong on the basis of this research

to side with those (e.g., Beer et al., 1984, p. 114; see Wood, 1996, for a

fuller discussion of the relationship between merit pay and high commit-

ment manage ment) who see merit pay as antithe tical to high commitment

manage ment. In fact, the use of merit pay for all was concentrate d in the

high HCM group, with all other occurrences of it falling within class 2. In

addition, the use of merit pay for some was, however, by no means con-

centrated in the high HCM category with nearly 50% of the workplace s

with some merit pay falling in the medium-high (third category) . The pro-

portion of workplace s with merit pay was close to 48% in both the third

and fourth classes. But the research does suggest that if the relative ly small

usage of merit pay for lower leve l participants is confirmed by further re-

search then its discriminatory power will still be low.

THE ASSOCIATION BETWEEN HCM AND NONUNIONISM

We first assessed the relationship between HCM and unionism by in-

cluding an indicator of unionism along with the other HCPs in a late nt

class model. We fitted up to four latent class mode ls to the data. The fits

with 3 and 4 classe s (Table VI) were both good, the latter mode l giving

substantially more information.

Comparing the results of the four class model with the union variable

include d with the one without it, we see that the freque ncies of the first

two categories were almost identical. The addition of the union recognition

item revealed that the first class is more like ly to be nonunion than un-

ionized, but organizations in it cannot be treated as definite ly not having


had trade unions. In contrast, one or more union is like ly to be recognized

by organizations in the second class.

This addition of the union variable to the analysis did, however, affect

the classification at the higher end of the scale, as the proportion of es-

tablishme nts in class four fell, with a corresponding increase in those in

class three. This change in the classification of establishme nts when union-

ism was include d in the analysis implie s that it remains a distinctive di-

mension on which to differentiate between organizations.

Second, ne ithe r high nor low HCM workplace s were distinctive in re-

spect to union recognition. Medium-high HCM workplace s were, however,

significantly less like ly to have one or more trade union recognized for

bargaining purpose s (only 17% of these have a recognize d union, which

compare d with an ave rage of 45% for the othe r groups) .

Third, orde red probit analysis (Table VII) revealed that union recog-

nition (or union density) did not have an inde pende nt effect on the leve l

of HCM controlling for othe r factors such as the size of the workplace and

whether the workplace is part of a wider organization or a single estab-

lishme nt organization.

These three analyse s suggest that in the U.K., HCM was not as of

1990 replacing unionism. They also confirm that low leve ls of HCM should

not necessarily be associate d with nonunionis m, as for example Sisson

(1993, p. 207) did when he labe led nonunion workplace s “bleak houses”.

THE RELATIVE PERFORMANCE OF HCM WORKPLACES

Within WIRS90, there are a number of performance indicators which

can be used to assess whether in fact high HCM establishme nts were more

Table VI. A 4-Class Model (Seven Items + Union): Prob-

abilities of Having a Practice a

Class 1 Class 2 Class 3 Class 4

1 0.95 0.96 0.94 0.95

2 0.49 0.82 0.84 0.873 0.18 0.34 0.29 0.80

4 0.24 0.11 0.13 1.005 0.08 0.02 1.00 0.98

6 0.50 0.64 0.99 1.007 0.37 0.96 0.89 0.92

8 0.39 0.83 0.65 0.19Size 0.19 0.49 0.28 0.04

aChi-square statistic = 150 (146 df), loglikelihood ratio statistic =

190 (146 df). Items: 1. internal recruitment or multiskilling or

training needs analysis, 2. direct communication, 3. appraisal, 4.

no clocking in, 5. monthly pay, 6. cashless pay, 7. information

disclosure, 8. union recognition.


effective than others. More specifically there are measure s of productivity,

job creation, labor turnove r, and abse nteeism. There is also a question on

the employe e relations climate which can be used as a measure of the in-

dustrial relations in the plant. We will briefly describe the construction of

these measures.

Productivity Level

This measure is based on answers to a question asking responde nts to

assess “how does the leve l of labor productivity” in their workplace “com-

pare with other similar workplace s? ” Responde nts were aske d to compare

it on a 5-point scale ranging from 1 equals a lot higher and 5 a lot lower,

with 3 about the same.

Productivity Change

This measure is base d on answe rs to the respondents’ assessment of

how the level of productivity in the ir workplace “compare s with what it

Table VII. Order-Probit Regressions on High Commitment Manageme nt with

Selected Variables

Dependent variable

Independant variables b t

1. Size 0.00 1.132. Size2

¯5.75 ¯0.70

3. Proportion of manual workers ¯1.14 ¯5.82***4. Proportion of part-time workers ¯0.44 ¯1.53

5. Proportion of female workers 0.42 1.80*6. Proportion of managers 0.16 0.24

7. Few or no competitors 0.22 2.41**8. One or more unions recognize d 0.03 0.02

9. Unemployment ¯0.06 ¯2.64***10. Qualified personnel manage r on site 3.00 3.00***

11. Gre enfield site 0.18 1.82*12. U.S. owned 1.29 5.06***

13. (E.E.C.) European-ownership 0.43 1.96*14. Non-E.E.C. European-ownership 0.35 0.99

15. Rest of world-ownership ¯0.11 ¯0.29Pseudo R2 0.14

n 899

*p £ 0.10.

**p £ 0.05.

***p £ 0.01.


was 3 years ago? ” Again the assessment was made on a similar 5-point

scale to the productivity leve l measure.

Finan cial Performance

This is a measure of the overall financial performance of the workplace

as judge d—relative to similar workplace s—by the responde nt on a 5-point

scale where 1 equals much worse and 5 equals much better.

Job Creation

This is a measure of the growth or decline in the employme nt leve l

ove r the 6 years prior to this survey (i.e ., 1984¯1990) .

Employee Relation s Climate

This is based on answe rs to a que stion which aske d responde nts to

rate “manage ment/employee relations generally at this workplace ?” on a

7-point scale thus: 1 equals very good, 2 unlabe led, 3 equals good, 4 un-

labe led, 5 equals poor, 6 unlabe led, and 7 equals very poor. Because of

the low frequencie s in the categorie s 5, 6, and 7, the scale was collapse d

into a 5-point scale by combining these three categories into one.

Labor Turnover

This is a ratio of the numbe r of employe es who resigned from the

establishme nt in the 12 months prior to the interview as a proportion of

the total employees at the time of the interview.

Absenteeism

This is the proportion of employe es at the workplace that “were away

sick or absent during the most recent period for which responde nts had

figures” (responde nts were also asked to indicate the period this covered).

The data for the productivity and financial performance questions were

large ly confined to the private sector, though some public trading organi-

zations were include d.

There are several weaknesses in these measures. First, the two pro-

ductivity measure s are base d on subjective assessments and one has to treat

the level of productivity measure with special caution because the majority


of workplace s were evaluating themselves as above ave rage (the mean for

the sample was 3.50). Second, the employee relations climate que stion is

in dange r of being a rathe r crude attempt to reduce a multidime nsional

issue to a unidime nsional scale . The variability in the response s was very

low with over 93% falling within eithe r the second or third categorie s,

meaning that the variance in response s is only accounting for the difference

between degrees of good climate . Third, the measure of absenteeism ap-

pears to be affected by the period over which the figures were compiled,

for the average abse nteeism rate was significantly diffe rent according to

whether the pe riod was a month or a week. Despite such proble ms we

report the results of an analysis of the relative performance of the four

types of HCM workplace s.

In orde r to assess whether the high HCM group did outperform the

others, we used regression mode ls to test the null hypothe sis that there

were no significant differences between the performance of the first three

classe s—low, low/medium, medium/low HCM—and high HCM. Table VIII

reports the results, which show that there was no tendency for the high

HCM workplace s to perform better than all others on any performance

criteria.

The high HCM group did, however, have a significantly greater leve l

of employme nt growth and ove rall financial performance than the two me-

dium categories, though it was not greater than the low HCM group. In

common, with much of the research (predominantly U.S.) , which has un-

covered performance gains from the use of HCM or other similar bundle s

of human resource practices, it is not easy to explain these results, but they

do suggest that HCM in a total form does have some effects but that these

are not unique to it. The possibility that low HCM (what Fernie & Metcalf,

1995 call authoritarianism or Sisson’s bleak house ) and high HCM are both

more successful, on some dimensions, than hybrid forms has been sugge sted

by the job creation and financial performance results.

Equally, though, the medium (low/medium and medium/high) HCM

groups did have significantly lower leve ls of abse nteeism than the high

HCM group, which is in fact one of the two measure s on which it might

be expe cted that HCM has more of a direct effect than financial involve -

ment or job creation.

The lack of any effect on variation in the employee relations climate

is understandable since the bulk of the variation to be explaine d is at one

end of the measure . But the lack of an e ffect on eithe r of productivity

measure s may in the end be most telling, as productivity is probably the

most significant measure of an organization ’s performance on the labor

front.


Table

VII

I.R

eg

ress

ion

so

nP

erf

orm

an

ceO

utc

om

es

wit

hS

ele

cte

dV

ari

ab

les

Pe

rfo

rma

nce

vari

ab

le(m

od

el)

Le

vel

of

lab

or

pro

du

ctiv

ity

(ord

ere

dp

rob

it)

Ch

an

ge

in

lab

or

pro

du

ctiv

ity

19

87¯9

0

(ord

ere

d

pro

bit

)

Fin

an

cia

l

pe

rfo

rma

nce

(ord

ere

d

pro

bit

)

Job

cre

ati

on

(OL

S)

Em

plo

yee

rela

tio

ns

clim

ate

(ord

ere

d

pro

bit

)

La

bo

r

turn

ove

r

(lo

gis

tic

reg

ress

ion

)

Ab

sen

tee

ism

(lo

gis

tic

reg

ress

ion

we

igh

ted

by

em

plo

yme

nt)

Ind

ep

en

da

nt

vari

ab

lesa

bt

bt

bt

bt

bt

bt

bt

1.

Lo

wH

CM

0.0

8¯0

.46

¯0.2

4¯1

.45

¯0.1

40

.71

¯0.3

5¯1

.45

0.1

91

.31

0.1

10

.86

0.1

31

.31

2.

Lo

w/m

ed

ium

HC

M0

.13

¯0.8

6¯0

.02

¯0.1

3¯0

.36

**

¯2.0

3¯0

.76

**

*¯3

.33

0.1

00

.77

0.1

21

.09

¯0.1

3*

¯1.7

83

.M

ed

ium

/hig

hH

CM

0.3

3*

1.9

60

.21

1.3

0¯0

.36

*¯1

.92

¯0.8

9*

**

¯3.7

9¯0

.15

¯1.0

8¯0

.03

¯0.2

5¯0

.45

**

*¯4

.76

4.

On

eo

rm

ore

un

ion

reco

gn

ize

d¯0

.30

**

*¯3

.11

0.0

50

.51

¯0.1

6¯1

.65

¯0.3

7*

**

¯2.7

9¯0

.18

**

¯2.2

5¯0

.33

**

*¯4

.44

0.0

60

.74

5.

Siz

e0

.00

0.8

70

.00

0.6

20

.00

0.4

0¯0

.00

¯0.0

50

.00

¯1.5

40

.01

¯1.7

40

.00

**

*4

.79

6.

Siz

e2

7.0

2¯0

.92

¯3.6

3¯0

.47

¯4.0

0¯0

.40

2.3

10

.02

4.0

70

.59

1.6

10

.72

¯4.1

1*

**

9.4

37

.F

ew

or

no

com

pe

tito

rs0

.29

3.3

00

.00

0.0

50

.07

0.7

6¯0

.42

**

*¯3

.30

¯0.2

6*

**

¯3.3

4¯0

.09

¯1.3

80

.18

**

*3

.65

8.

Un

em

plo

yme

nt

0.0

3¯1

.66

0.0

10

.33

¯0.0

2¯0

.67

0.0

31

.01

¯0.0

3¯1

.44

¯0.0

5*

**

¯2.7

50

.03

**

*3

.80

R2

0.0

70

.07

0.0

70

.12

0.0

60

.27

0.3

4n

85

19

26

80

67

94

12

41

10

70

11

17

aT

he

eq

ua

tio

ns

als

oin

clu

de

dth

efo

llo

win

gco

ntr

ol

vari

ab

les:

cert

ain

wo

rkfo

rce

cha

ract

eri

stic

s(e

.g.,

%p

rop

ort

ion

of

ma

na

ge

rs)

an

din

du

stry

du

m-

mie

s.S

ign

ific

an

tly

hig

he

r/lo

we

rth

an

hig

hg

rou

p:

*0

.05

£p

<0

.10

;*

*0

.01

£p

<0

.05

;*

**

p<

0.0

1.


CONCLUSION

The starting point of our analysis was to treat HCM as a matter of

degree. This contrasts with many previous writers, who have tende d to focus

on high commitment practice s as a total package and on discovering low

adoption of individual HCPs inferred a fragmented approach to HCM. In

this paper, we have demonstrated that there is coherency to the use of

high commitment practices. A latent class model which treats the under-

lying dimension as discre te could be fitted to the data. As such high com-

mitment manage ment exists, and we have discove red four progre ssive types

of HCM.

The implication of our results is that Sisson’s and othe rs’ discovery of

fragments of HCM is not a reflection of manage ments’ ad hocism. Nor

doe s it seem to be a result or artifact of the research instrume nt, the survey

method, or a lack of sufficie nt measures of HCM. It is large ly a result of

the type of analyse s which they conducte d on WIRS90, based as they were

on freque ncies and cross-tabulations. We cannot, however, rule out the pos-

sibility that with the inclusion of the other elements of high commitment

manage ment—concerning job flexibility, job security, and teamworking—a

continuous late nt variable could successfully be fitted. Equally, though,

analysis with any additional items may reinforce the basic story of this

analysis.

The study suggests that the numbe r of organizations with medium or

high leve ls of high commitment management—whether this is treated as

taking one form or not—are probably more prevalent than writers who

treat HCM as an absolute concept imply. Certainly, given the high fre-

que ncy of workplace s falling in the two medium groups, the kind of or-

ganization which Sisson (1993, 1995, p. 106) saw as the “bleak house”typical of small and medium-sized non-union firms, is far from being the

majority. Moreover, we have shown that the low HCM class, though seem-

ingly corresponding to Sisson’s bleak house , is not necessarily nonunion.

More generally, the lack of a strong relationship between union recognition

and HCM is consistent with other related research (e.g., Lawler et al., 1995;

Osterman, 1994; Wood & Albane se, 1995) .

This research has also shown that the provision of welfare facilitie s is

a separate dimension on which we might differentiate between workplace s.

Such a conclusion implie s that we need to clearly separate many of the

umbrella terms that have been used to describe the “new” organization or

employment relations and be particularly mindful of terms such as the car-

ing organization, which are often used interchangeably with high commit-

ment manage ment.

Caution has to be exercised in making conclusions from the analysis

linking HCM to performance, but it revealed no unique strong performance


gains stemming from its use . It doe s appear that exceptionally high users,

in common with low users of HCM, do perform better than other organi-

zations in terms of their profitability and ability to create jobs. Nonetheless,

on othe r dimensions, the results are consiste nt with those of Edwards

(1995) who conclude d HCM had little effect. However, they do not support

his conje cture (ibid. p. 218) that this lack of association may well be a re-

flection of the piecemeal way HCM is being adopte d, since, as we have

said, these results are not consistent with the ad hoc thesis of HCM.

It is difficult to compare the results of this study with others, since it

is the first of its kind to use late nt trait analysis on binary data on high

commitment practices from a nationally representative survey of workplace s

which also contains performance data. But the study by Guest and Hoque

(1994) , which was able to link high commitment practices to performance

outcomes, revealed a similar patte rn of uneven results across diffe rent per-

formance measure s, even though its authors conclude d that ove rall the ir

systematic use did appear to have beneficial effects.

The results are , however, less consiste nt with the U.S. studies, the ma-

jority of which are at the industry level and moreove r conce ntrated in

manufacturing. As Ichniowski et al. (1996) show in the ir review, most stud-

ies which examine systems of practices have demonstrate d that they are

associated with positive performance effects.

Given the dive rsity of performance variable s, measure s of performance

and individual practices and modes of combining them into bundle s to dif-

ferentiate between organizations, it is difficult to assess whether the diffe r-

ence be tween the results of this study and those dominating the U.S.

research reflects a basic diffe rence between the U.S. and U.K. It may be

more a reflection of the different sets of items being used and/or outcome

variable s. There are important diffe rences between our study and the most

significant national study in the U.S. by Huse lid (1995) which make it es-

pecially difficult to compare the results. In particular, his study is at the

company level and this is at the establishme nt, his performance data are

less subje ctive , and perhaps the link between his (input) measure of the

bundle of practice s to high commitment management is more tenuous than

is the one deve loped in this study.

It may yet be that the industry level is the most appropriate , though

there seems no reason why positive results at the industry-le vel should not

be reproduced nationally. Moreover, since the U.S. industry-le vel results

are all in manufacturing we tested to see if the performance effects of HCM

in Britain we re unique ly positive whe n manufacturing was conside red

alone ; the results were no different.

Two possibilitie s cannot however be ruled out at this stage ; that with

better performance variable s, there might be a positive association between


HCM (as measured in this study) and performance, or alte rnative ly were

the same measure of HCM used in the U.S., with or without better per-

formance measures, we might get the same results.

We certainly cannot conclude that HCM doe s not work from this

analysis. First, as continge ncy theory argues, HCM may not be universally

applicable and hence will only have strong effects on performance in un-

stable environme nts. Accordingly—assuming that both continge ncy theory

is right and that the high performers have all successfully fitted the ir human

resource manage ment to their environme nt—our results could be a reflec-

tion of the fact that some of the high performers are ope rating in a stable

environme nt.

Because of a lack of good data on contextual variable s such as the

competitive strategy of the organization, we were unable to test the con-

tingency thesis that HCM is only positive ly related to performance in cer-

tain environme nts. Analysis of the industrial distribution (available from

the authors) does not sugge st that the use of HCM is concentrated in only

a small numbe r of the more “dynamic” or science-based industrie s, banking

and finance be ing the one sector where the proportion of workplace s in

the high HCM category was significantly greater than that for the vast ma-

jority of othe r industrie s.

Second, there are a number of links in the chain running from the

use of high commitment manage ment to performance , and it may be that

it is not having any significant effect on employees’ commitment, or if it

is, this is not be ing translate d to tangible behaviors. Or it may be that HCM

impacts, at least in the short-run, on criteria othe r than those we have

measure d, such as the achievement of quality standards and the speed of

introduction of change .

Third, it is possible that the omission of job security guarante es and

job design principle s has affected our results. This may mean that we have

not ade quate ly distinguishe d between those organizations which treat high

commitment management as a process of making mutual obligations from

those which use high commitment in a large ly instrumental way, that is as

a one -way process concerned with eliciting commitment from employe es.

In addition, it could yet be that the benefits of HCM are only realizable

where obligations are mutual.

Despite these pote ntial reasons for the lack of a strong association

between HCM and many of the performance variable s, we clearly cannot

rule out the possibility that there is no unique pay-off to using HCM, that

is that the results can be take n at face value . The implication of this would

be that argume nts for HCM have to be centered on equity rather than

efficiency grounds.


Finally, the research has a number of broader implications. First, though

the results show that high commitment manage ment is a potentially useful

way of differentiating between organizations, they imply that it is not a ques-

tion of replacing unionism with high commitment manage ment as the main

axis on which to divide organizations for employee relations purpose s. Second,

the success of our analysis might be used to support the basic approach of

surveys such as the Workplace Industrial Relations Serie s (WIRS); overcom-

ing the deficiencies in the data on HCM should not then amount to a radical

change in the orientation or form of the survey. In short, such surveys can

usefully remain centered on institutionalize d practices. In the specific case of

WIRS, it should continue to extend its coverage of areas of personnel man-

agement conceived by its original archite cts as marginal to employe e relations

and incorporate some of the questions and lessons learned from other studies,

including EMSPS. (Though more questions should be included on the organ-

izational and environmental context in which employee relations systems are

embedde d, and additions and improve ments to the performance variable s

could be made). Finally, in showing that management may not be approaching

personnel manageme nt in a piecemeal way, this research questions the influ-

ential stereotype of British management which assumes the pervasiveness of

reactive and segmented approache s to all areas of business.

ACKNOWLEDGMENTS

The study was conducted as part of the ESRC Analysis of Large and

Complex Datase ts Programme (ALCD) at the Methodology Institute , Lon-

don School of Economics and Political Science. During the research Stephen

Wood held an ESRC ALCD Research Fellowship. The financial support of

the ESRC is gratefully acknowle dged. We would also like to thank Professors

David Bartholome w, Paul Marginson, David Metcalf, and Toby Wall and two

anonymous referees for their comments on earlie r drafts of this paper.

The research has also benefitted from discussions with Dr Neil Mill-

ward and Stephen Woodland, especially on the nature of the Workplace

Industrial Relations Survey. In addition, we would like to thank Stephen

Woodland for supplying the abse nteeism and labor turnove r equations re-

ported in Table VIII.

Data from this study were supplie d by the ESRC Data Archive at Es-

sex Unive rsity in Britain, whose assistance is grate fully acknowle dged. The

Workplace Industrial Relations Surve y data were deposite d in the Archive

by its joint sponsors, the Employment Department, the Economic and So-

cial Research Council, the Policy Studie s Institute , the Advisory, Concili-

ation and Arbitration Service . The E mploye rs Manpower and Skills

Practices Survey was deposite d by its sponsors, the Employment Depart-


ment. Neither the de positors nor the ESRC Data Archive have any re -

sponsibility for the analysis or interpretation of the material containe d in

this paper.

APPENDIX

Laten t Variab le Models

Latent Variable mode ls (as in Bartholome w, 1987, Chaps. 2, 4, and 5)

are used to assess whether a set of obse rved variable s or indicators are

linked by an underlying variable which is not readily observable . In statis-

tical terms, the association between these obse rved or manife st variable s

is explaine d by one or more common factor(s), and they are assumed to

be inde pendent but for this underlying factor. Latent variable mode ls can

be used for two relate d purpose s: to identify and deve lop measures of such

unde rlying factors as for example abilitie s and orientations, and to assess

whether a set of items thought to form a unity do in fact co-exist.

We can distinguish four type s of late nt variable analysis: (1) factor

analysis, when both the manife st and late nt variable s are continuous, (2)

late nt trait analysis, when the observed or manife st variable s are categorical

(binary) and the underlying latent variable is assumed to be continuous,

(3) latent class analysis, when both the latent and manife st variable s are

categorical (binary inclusive ), and (4) latent profile analysis, when the mani-

fest is continuous and the latent variable is categorical.

In our study, we use a latent trait model (logit-probit) and standard la-

tent class model, for which we use the software develope d at the London

School of Economics (Department of Statistics and Methodology Insti-

tute—ALCD Project). The items are high commitment practices and the re-

sponde nt has a choice of either indicating whether the establishme nt for

which information is be ing sought has a particular practice , or of not answer-

ing. His or her response can be summarize d as a set of 1s when the estab-

lishment does have an item, 0 when the establishme nt does not, and 9 if the re

is no answer (i.e ., a missing case). The set of his/he r response s is thus called

a response patte rn. For example , if there are four HCPs under investigation,

then the response patterns can range from 0000 to 1111, while 9999 would

mean that the respondent had not answered any of the respective questions.

Using the notation and terminology of Bartholome w (1987, pp. 107-

129) , the response pattern consists of p binary observable variable s (ques-

tions or ite ms) denote d by x1,...,xp and can be e xpre sse d for the sth

individual thus: xs= (xis,... xps) where xis,x2s,... xps are p variable s taking values

0, 1, or 9, and xis is the value of the ith variable xi = 1 for the sth estab-


lishme nt, s = 1,...n, xis = 1 indicate s that the establishme nt has the HCP

and xis = 0 indicate s that it does not have it.

The models which we are using assume that the answe r to a particular

item is independent of the answers provide d for any othe r item give n the

late nt variable (e.g., the underlying attitude toward high commitment man-

agement). This corresponds to the assumption that the latent variable (s)

explains all the association between the answe rs to diffe rent items for a

particular establishme nt.

The Latent Trait Model

We consider a two-factor logit-probit model; the probability of a positive

response for the ith item (having the ith HCP) is supposed to depend on the

two continuous latent variable s and denoted , where z = (z1, z2) are the

late nt variable s. If the association among the xs, or HCPs, are totally ac-

counted for by their common dependence on the latent variable (s), then the

xs are independent. Hence, give n z, xi is a random variable with probability

function:

For reasons give n in Bartholome w (1987) , the response function is specified

as

The a parameters determine the shape of the response function and are

estimated by marginal maximum like lihood. The extremeness of an item is

measure d by which is known as the difficulty parame ter. The parameter

(j= 1,2) are referred to as the discrimination parame ters because as co-

efficients of z1 and z2, the ir size determines the effect which a give n change

in z has on the probability of having a particular practice . For two estab-

lishme nts a give n distance apart on the z scale, the bigge r the absolute

value of the greater the diffe rence in the probabilitie s of their having

a practice and thus the more like ly we are able to discriminate between

them on the basis of their answe rs to item i. The discrimination parameter

( ) may be thought of as equivale nt to the factor loadings or weights in

the linear factor analysis. A better link to factor analysis is then achieved

by the following standardization

p i ( )z

P xi ixi

ixi xi i p( | ) [ ( [ ( ( , ,... ,z z z= - = =p p)] )]

1-or 1; )1 0 1 2

logit { ip a a a( )} ( ,. .. , )z = + + =i i z i z i p0 1 1 2 21

a i0a ij

a ij

a ij

a aij ij A*

/=


and j = 1,2. The probability that an individual (establishme nt), which is at

the median point of the latent scale s, responds positive ly (i.e ., has the ith

HCP) is

The late nt variable s (z) are conve ntionally assumed to be standard normal

variable s. Bartholome w (1987) has argued that the form of the prior dis-

tribution is relative ly unimportant and thus the choice of the normal dis-

tribution is merely for convenience.

The above mode l assumes that every individual has the same prob-

ability of answe ring a given item, and that not answe ring (9) give s no in-

formation about the latent variable . For each item the probability of not

answering is assumed constant ove r all individuals, independently of the ir

answers to other items.

If the fit of the latent variable model to the data is good, we have late nt

variable scores for each establishment. Rather than simply aggre gating the

number of positive response s to the set of items, we can use the expected

value of the late nt variable after the response pattern has been observed as

a score. In the present study, if we fitted a one-factor mode l (z2 = 0), we

could then consider the scores or value s of z1 as a measure of high commit-

ment management.

Latent Class Models

Following the notation used in Bartholome w (1987, pp. 17-27) , the

classical late nt class mode l is summarize d be low.

Suppose there are p binary variables x1,..., xp with xi = 0 or 1 for all i, and that

their association is due to discrete points (classes) in the latent space. Let ( j) bethe probability of a positive response on variable i for an individual in cate gory j

and let be the prior distribution that a randomly chosen individual is in classj. As only the manifest ve ctor, x, can be ob se rve d, in fe r e nce m u st be base d

on the joint distribution whose density for the K classes mode l becomes

The probability of an individual (establishment) belonging to a class j give n his/her

(its) responses to the questions (x) is therefore

Ai i

= + + + -( )/

11

22

2 1 2a a

pai

e iz=

+ -= = =

1

1 01 0

( )|Pr(xi )

p i

h j

f

j

K

jt

p

i jxi

i jxi( ) ( ) ( ( ) )x =

=

-

=- -å Õ

0

1

1

11h p p


The above value is known as the posterior probability and is used as the basis for

allocating individuals into different classes. For example, a workplace can be placed

in the class (category) for which the posterior probability is the greate st. Similarly

to scale-measures or scores, these classes can then be used for further analysis.

In the present study, if we fitted a K-class mode l to the data, then K ho-

mogeneous groups of establishme nts would be identified on the basis of

the ir style s of approaching high commitment manage ment. As in cluster

analysis, groups of establishme nts would be located; however, in this case ,

these groupings are based on the ir position on the late nt scale .

Parame ters of the above mode ls are estimated by maximum like lihood,

for which the E¯M algorithm (Dempster, Laird, & Rubin, 1977) has been

traditionally applie d.

Goodness-of-Fit of the Models

In the present study, we have used three methods for judging the good-

ness-of-fit of both the mode ls. The first is to compare the obse rved and

expected freque ncies for each of the 2p response patterns using the log-

like lihood ratio statistic (G 2), which is given by

where Oi and Ei are the obse rved and expe cted frequencies for the ith

response patte rn.

The above statistic approximates a chi-square distribution with 2p-2p-1 de-

grees of freedom for a one-factor model and 2p-3p-1 degrees of freedom for a

two-factor model. However, as the number of response items becomes larger

(p > 7), expected frequencies become smaller and response patterns have to

be grouped in order to have large enough expected frequencies to justify the

chi-square approximation. At each grouping, one degree of freedom is lost, and

when many groupings are required the test may no longer be feasible .

The second method involve s looking at the reduction achieved in the value

of the log like lihood by fitting the model. That is, if we denote by H0 the hy-

pothesis that the items are independent and by H1 the latent variable model,

and let G2(H1) denote the log likelihood ratio statistic when H1 is true, then

h j ji

p

i jxi i j

xi f j K( | ) ( ) ( ( )) / ( ) ( , ,... , )x x==

- - = -Õh p p1

11

0 1 1

G O O Ei i i2

2= å log ( / )

[ ( ) ( )]

( )

G H G H

G H

21

20

20

100- ´


may be interpreted as a measure of the extent (percentage) to which the

original departure from independe nce is accounted for in the latent vari-

able mode l.

The third method is to compare the obse rved and expected two- and

three-way cross-tabulations and compute the discrepancie s between them.

We compare the actual number of responses (O) to a pair or a triple t of

items, (i,j) or (i,j,k), with its expected value (E) and using the measure

(O¯E)2/E, we assess how well the model has predicted a significant part

of the correlation in the data. Small value s, say less than 4, indicate a good

fit. However, this method is an informal goodne ss-of-fit criterion, since the

value s cannot be added to form a valid “chi-square d test.”

REFERENCES

APPELBAUM, E., & BATT, R. The new American workplace: Transform ing work systems in

the United States. Ithaca, NY: Cornell IR Press, 1994.

BARTHO LOMEW, D. Latent variable models and factor analysis. London: Charles Griffin,1987.

BEAUMO NT, P. The future of em ploym ent relation s. London and Thousand Oaks: Sage, 1995.

BEER, M., SPECTOR, B., LAWRENCE , P., MILLS, D., AND WALTON, R. Managing hu-

man assets. Ne w York: The Free Press, 1984.

BOXALL, P. F. Strategic human resource manage ment: Beginnings of a new theoretical so-

phistication? Hum an Resource Managem ent Journal, 1992, 2(3), 60-79.

DEMPSTE R, A. P., LAIRD, N. M., & RUBIN, D. B. Maximum likelihood from incomplete

data via the EM algorithm. Journ al of the Royal Statistical Society, Series B, 1997, 39(1) ,1-22.

EDWARDS, P. K. Human resource management , union voice and the use of discipline: An

analysis of WIRS3. Industrial Relations Journal, 1995, 26(3), 204-220.

EMPLOYERS’ MANPO WER AND SKILLS PRACTICES SURVEY. Docum entation for usewith Machine-Readable Data, Colchester: ESRC Data Archive, Unive rsity of Essex, Col-

cheste r, 1994.

FERNIE, S., & METCALF, D. Participation, contingent pay, represe ntation and workplaceperformance : evidence from Great Britain. British Journal of Industrial Relations, 1995,

33(3) , 379-415.

FERNIE, S., METCALF, D., & WOODLAND, S. Lost your voice? New Econom y, 1994, 1(4) ,

231-237.

GUEST, D. Human resource manage ment, trade unions and industrial relations. In J. Storey

(Ed.), Hum an resource m anagement. London and New York: Routledge, 1995, pp. 110-141.

GUEST, D. Human resource manage ment and industrial re lations. Journal of Managem ent

Studies, 1987, 24(5) , 503-521.

GUEST, D., & HOQUE, K. The good, the bad and the ugly: Employme nt in non-union

greenfield sites. Hum an Resource Managem ent Journal, 1994, 5(1), 1-14.

HUSELID, M. The impact of human resource management practices on turnover, produc-

tivity, and corporate financial performance . Academ y of Managem ent Journal, 1995, 38(3) ,635-672.

ICHNIO WSKI, C., KOCHAN, T., LEV INE, D., OLSON, C., & STRAUSS, G. What Worksat Work: Ove rview and Assessment. Industrial Relations, 1996, 35(3) , 299-333.

KALLE BERG, A. L., KNO KE, D., MARSDEN, P., & SPAETH, J. Organizations in America.

Thousand Oaks: Sage, 1996.

KOCHAN, T., & OSTERMAN, P. The Mutual G ains Enterprise, Cambridge , MA: Harvard

Business School Press, 1994.


KOIKE, K. Human resource deve lopment and labour-management relations. In Yamamura,

K., & Yasuba, Y. (Eds.) The Political Econ omy of Japan (Vol. 1), The Dom estic Transfor-mation, Stanford: Stanford Unive rsity Press, 1987, 289-330.

LAWLER, E. High involvem ent m anagem ent, San Francisco: Jossey-Bass, 1986.LAWLER, E., MOHRMAN, S., & LEDFO RD, G. R. Creating high perform ance organizations.

San Francisco, Jossey-Bass, 1995.LAZARSFELD, P. F., & HENRY, N. W. Latent structure analysis. Boston: Houghton Mifflin,

1968.

MILLWARD, N., STEV ENS, M., SMART, D., & HAWES, W. R. Workplace industrial relationsin transition. Aldershot: Dartmouth, 1992.

MILLWARD, N. The new industrial relations. London: PSI Publishing, 1994.OSTERMAN, P. How common is workplace transformation and who adopts it. Industrial Re-

lations and Labour Relations Review, 1994, 47(2) , 173-188.SISSON, K. In Search of HRM. British Journal of Industrial Relations, 1993, 31(2), 201-210.

SISSON, K. Human resource managemen t and the personnel function. In J. Storey (Ed.),Hum an resource managem ent. London: Routledge , 1995, pp. 87-109.

STOREY, J. Developm ents in the Managem ent of Hum an Resources. Oxford: Basil Blackwell,1992.

STOREY, J. Human resource managemen t: Still marching on, or marching out? In J. Storey(Ed.), Hum an resource managem en t. London: Routledge, 1995, pp. 3-32.

WALTON, R. From “control” to “commitment” in the workplace. Harvard Business Review,1985, 63(2), 77-84.

WOOD, S. High commitment manage ment and payment systems. Journ al of Man agem ent Stud-ies, 1996, 33(1) , 53-77.

WOOD, S. How different are human resource practices in Japane se “Transplants” in the UK?

Industrial Relations, 1996, 35(4) , 511-525.WOOD, S., & ALBANESE, M. Can you speak of a high commitment manage ment on the

shop floor? Journal of Managem ent Studies, 1995, 32(2), 215-247.

BIOGRAPHICAL NOTES

STEPHEN WOOD is Reader in Industrial Relations at the London School of Economics

and Political Sciences, and Editor and Reviews Editor of the British Journal of Industrial Re-lations.

LILIAN DE MENE ZES is a Lecturer in Statistics at Goldsmiths College, Unive rsity of Lon-

don. She obtained her PhD from the London Business School. He r current research interestsinclude forecasting, latent variable models, and measureme nt in the social sciences.


High Commitment Management in the U.K.: Evidence from the Workplace Industrial Relations Survey, and...

Documents

Transcript of High Commitment Management in the U.K.: Evidence from the Workplace Industrial Relations Survey, and...